That compression is not a retail phenomenon. It is a structural shift in how skincare innovation moves through the world. And it is accelerating. 

South Korea is the world’s No. 2 cosmetics exporter as of Q1 2025, with $3.61 billion in shipments, trailing only the United States ($3.75B). US imports of Korean cosmetics reached $1.7 billion in 2024, a 54% year-over-year surge, while Japan’s K-beauty imports jumped 120.8% that year, surpassing French imports. 

This makes K-beauty the global gold standard for skincare innovation.

Korean beauty trends are notoriously known for their fast lifecycle, which makes many R&D leaders outside Korea hesitate about which ones to pick for the next R&D cycle. 

This report identifies five transformative trends that are reshaping the Korean beauty landscape in 2026. 

Trend 1: MediCosmetic: PDRN

Medicosmetic describes the rapid migration of active ingredients once exclusive to clinical aesthetic medicine into over-the-counter, daily-use Korean skincare products sold at accessible price points on Olive Young and Coupang. These ingredients include PDRN (polydeoxyribonucleotide), EGF (epidermal growth factor), tranexamic acid, dexpanthenol, and exosome-infused topicals.

PDRN, originally developed for wound healing and first used in Korean aesthetic clinics via Rejuran injectable boosters around 2014, made its mainstream retail debut in 2017 with Rejuran’s skincare line.

However, the true acceleration came in 2023, when Jennifer Aniston referenced the “salmon sperm facial” in a Wall Street Journal interview. And later, Kim Kardashian showcased getting a PDRN facial on The Kardashians, catalyzing global consumer curiosity.

On Olive Young, PDRN appears in 33 of the Top 50 skin longevity products, behind only Hyaluronic Acid (35) between October 2025 and March 2026.

Three structural forces make this trend durable:

• Consumer shift toward clinically proven efficacy over aspirational “glow” marketing, particularly among post-pandemic consumers who became comfortable with at-home clinical tools.
• TikTok’s K-Pharmacy pipeline: OTC K-pharmacy brands (Rejuran, IOPE, VT Cosmetics) that were once Korea-only are now viral sellouts at Olive Young’s global e-store within days of going live.
• Price democratization: Where in-clinic PDRN facials in Seoul cost KRW 200,000–400,000 per session (~$150–300), OTC products like Medicube’s PDRN serum retail at $25–40, creating a 6–10x accessibility gap that fuels demand.

PharmaResearch Corp, a South Korean regenerative medicine and aesthetics company, has become a strong example of how PDRN and polynucleotide-based skin repair technologies are moving from clinical use into the cosmetic market.

The company works around patented DOT PDRN and DOT PN technologies, which are DNA-derived materials used for tissue regeneration. Its best-known brand, Rejuran, applies this platform to skin rejuvenation, especially through PN-based injectable skin boosters and PDRN-based cosmetic skincare products.

Rejuran uses purified DNA fragments, commonly sourced from salmon DNA, to support skin repair, hydration, elasticity, and anti-aging positioning. PharmaResearch’s Rejuran i product, for example, lists PN as its key ingredient and is intended for use on the face, eye area, neck, hands, and décolleté.

The company is bringing PDRN-linked solutions to market through two main routes: professional aesthetic treatments and consumer skincare. 

On the clinical side, Rejuran is marketed as a wrinkle-reduction medical device based on DOT PN technology, with PharmaResearch stating that it has been used by more than 500,000 people annually since its 2014 launch. On the skincare side, Rejuran Cosmetics uses patented c-PDRN® and positions it as a daily derma-skincare solution inspired by regenerative science.

This makes PharmaResearch a useful case study for the cosmetic PDRN trend: it is not only researching DNA-based regenerative ingredients but also converting them into branded, scalable products across clinics and skincare retail.

Trend 2: Biotech Actives: Exosomes

Exosomes are nano-sized extracellular vesicles (30–150 nm) that carry growth factors, microRNAs, and proteins. It functions as a skin’s cellular communication network, signaling repair, regeneration, and collagen synthesis. 

Unlike PDRN (which is a structural DNA fragment), exosomes deliver biological instructions directly to target cells.

Korean biotech labs began standardizing exosome protocols for clinical use in 2023. By 2026, the technology had moved from experimental to mainstream in Seoul dermatology clinics and was rapidly entering the retail market. 

Key drivers of momentum:

• Cell-level intelligence: Exosomes do not simply hydrate — they deliver instructions, stimulating fibroblast activity, collagen synthesis, and skin repair without the inflammatory response that retinoids or acids trigger. This “smart skincare” positioning resonates with a skin-barrier-conscious generation.
• Synergistic pairing: Seoul dermatologists are now combining PDRN (repair) + exosome (communication) in protocols, with brands rapidly translating this dual-active approach into serum formats. Products at this intersection command premium pricing ($25–$70 for OTC formats).
• Patent acceleration: KIPO filings around exosome-based delivery systems jumped ~60% from 2023 to 2024, with applicants including Amorepacific Research Institute, Huons (pharma), CHA Biotech, and multiple university spinouts. Several of these remain in development, with no commercial product yet, representing white-space licensing opportunities.

Under APR Corp, Medicube has built an exosome device‑serum ecosystem around products such as the Zero/One Day Exosome Shot and the PDRN Pink Collagen Exosome Shot. 

These serums use micro‑sized, needle‑shaped spicules coated with exosomes to physically enhance penetration, marketed as “microneedling in a bottle” and supported by claims of significantly higher absorption than with conventional application. 

Independent reviews note that Medicube’s exosome shot leverages patented lactobacillus‑derived exosomes carrying peptides, amino acids, glutathione, and vitamins, aligning tightly with the “smart vesicle” narrative in your trend write‑up.

On the peptide side, the PDRN Pink Collagen Exosome Shot combines high‑purity salmon PDRN with collagen, hyaluronic acid, niacinamide, and peptide complexes, marketed for plumping, wrinkle‑smoothing, and texture refinement, with strong social and Amazon traction across both 2,000 and 7,500 “exosome” versions. 

Strategically, APR is doubling down on biotech by bringing PDRN production in‑house and announcing plans for injectable skin‑regeneration therapies, blurring the line between clinic‑grade actives and cosmetic retail and reinforcing the medicosmetic–biotech continuum behind this trend.

Cosmetic Industry Trends 2026 Report

Download Report

Trend 3: Novel Korean Botanicals

While centella asiatica, snail mucin, and ferments have graduated to global commodity status, a new wave of distinctly Korean-origin botanicals is filling the innovation pipeline.

Heartleaf (Houttuynia cordata, known in Korea as eoseongcho, 어성초), a perennial herb native to the damp, shaded forests of Korea and East Asia, has been used in traditional Korean medicine for over 1,000 years to treat infections and inflammation.

Heartleaf’s commercial breakthrough came through Anua’s Heartleaf 77% Soothing Toner, which went viral on TikTok and became the #1 toner on Olive Young’s Glowpick rankings in 2024–2025, retailing at $18 for 250ml. 

As of September 2025, over 600 products in the Korean market list heartleaf as a key INCI ingredient, a 2X growth in 18 months, which confirms its evolution from a trend to a market standard.

Abib dominates with 13 heartleaf-centric products, while Anua deploys 11 across the full category range.

Beyond heartleaf, other novel Korean botanicals gaining traction include:

• Korean mugwort (Artemisia princeps) — An adaptogenic herb with antioxidant and anti-inflammatory properties; featured by Mixsoon and Beauty of Joseon.
• Jeju Island botanicals (green tea, volcanic minerals, tangerine extract) — These are regionally sourced, premium ingredients that command 15–20% price uplift.
• Rice ferment filtrates (not standard bifida) — A new proprietary strain of rice-fermented actives from small Korean biotech labs claiming superior skin-brightening vs. niacinamide.
• Spicule actives (sea sponge microcrystalline structures) — It is used by VT Cosmetics’ Reedle Shot line for at-home microchanneling effects; a novel delivery mechanism rather than a botanical, but uniquely Korean-commercialized.

Anua’s flagship Heartleaf 77% Soothing Toner is built around a very high heartleaf extract load and is positioned for soothing, hydration, redness reduction, and acne‑prone skin support, which fits the “post‑centella” shift toward next‑generation Korean calming botanicals.

On Olive Young, the product is sold as a 2025 Awards item, carries a 4.8 rating from hundreds of reviews, and is described as clinically tested for 24‑hour hydration, non‑comedogenicity, and reductions in acne grading score and sebum after four weeks.

What makes Anua especially relevant is not just one hero SKU, but how it has expanded heartleaf into a broader ingredient franchise. 

Olive Young’s Anua brand page explicitly defines the brand around naturally derived and derma ingredients “from Heartleaf for soothing the skin,” while listing multiple heartleaf products such as the Heartleaf Cleansing Oil, Heartleaf 77 Hyaluron Soothing Toner, Heartleaf Clear Pad, Heartleaf Quercetinol Pore Deep Cleansing Foam, and Heartleaf Silky Moisture Sun Cream.

This is exactly the commercialization pattern highlighted in your document, where heartleaf evolves from a single breakout botanical into a full routine architecture.

In strategic terms, Anua shows how Korean brands are converting local botanical heritage into globally legible problem‑solution skincare. Rather than selling “natural beauty” broadly, Anua frames heartleaf as a clinically supported calming active for sensitive, oily, and acne‑prone consumers.

Trend 4: Postbiotic & Microbiome Science

Korea’s fermentation tradition, which gave the world galactomyces ferment filtrate (SKII’s Pitera predecessor), bifida ferment lysate (Estée Lauder’s Advanced Night Repair backbone), and saccharomyces ferment, is now undergoing a second-generation revolution. 

The new frontier is postbiotics — the specific bioactive metabolites, enzymes, lipoteichoic acids, and exopolysaccharides produced by proprietary fermentation strains — rather than the live organisms (probiotics) or their food (prebiotics).

The global microbiome cosmetics market was valued at $631 million in 2024 and is projected to reach $1.24 billion by 2032 (CAGR 8.8%), with Asia-Pacific commanding the largest volume share (38%). 

Korea’s unique biodiversity, particularly microbial strains from the volcanic hot springs of Jeju Island and mountain forest soils, is yielding novel fermentation strains with skin-conditioning profiles that European or American competitors have not replicated.

Critically, Google searches for ‘skin microbiome’ rose 176.9% year-over-year between August 2024 and July 2025. 

The MFDS (Korea’s FDA) has begun issuing guidance on postbiotic cosmetic claims, indicating that the category is transitioning from niche to regulated mainstream status.

• Postbiotic market trigger: Next-generation postbiotic complexes demonstrated up to 35% improvements in skin barrier integrity compared with standard moisturizers in controlled studies, providing the clinical narrative that drives formulators and dermatologists.
• Beyond bifida: Korean biotech startups are filing IP around novel Lactococcus, Leuconostoc, and Weissella strains fermented in rice bran, sake lees, and endemic Korean botanicals — generating unique metabolite profiles not available from existing ingredient suppliers.
• Microbiome diagnostics convergence: Korean labs (BIOTALIFE, Sequential Korea) are coupling DNA sequencing of individual skin microbiomes with personalized postbiotic recommendations, creating a data-driven prescription skincare market estimated to reach $2.14 billion globally by 2033.

Ma:nyo’s Bifida Biome Complex Ampoule is explicitly positioned around skin-barrier strengthening, moisture retention, and wrinkle minimization, which aligns closely with postbiotic microbiome science as a barrier-first, clinically legible skincare trend. 

On its official US product page, Ma:nyo says the serum fortifies the skin barrier and includes “Bifida Biome,” five probiotics, and ten types of hyaluronic acid, while citing a three-week human study and consumer-reported calming benefits for skin damaged by external irritants.

Ma:nyo’s Bifida Biome line spans toner, ampoule, cream, pads, masks, and sets, demonstrating how a fermented/postbiotic ingredient system can be developed into a routine-based franchise rather than a one-off launch. 

The brand’s own ingredient explainer defines Bifida Biome as a combination of bifida ferment lysate and microbiome-supportive components intended to strengthen, restore, and replenish the skin barrier, directly matching the report’s argument that Korean brands are moving beyond basic bifida into more systematized postbiotic complexes.

Strategically, Ma:nyo shows how K-beauty brands are commercializing microbiome science in a consumer-friendly way. Instead of emphasizing technical strain names alone, Ma:nyo turns fermentation science into a clear value proposition: better barrier resilience, less irritation, and long-term skin health for sensitive or over-treated skin.

Trend 5: AI-Powered Personalization & Daily BeautyTech

This trend integrates two parallel Korean innovations, i.e., AI-driven skin diagnostics and daily-use smart-home beauty devices. Together, they represent the shift from product-as-commodity to skincare-as-a-service.

On the AI diagnostics side, Amorepacific has led the charge by partnering with Microsoft Azure OpenAI in March 2025 to launch an AI Beauty Counselor (AIBC) on Amore Mall. It offers personalized skin analysis and product recommendations based on purchase history and chat data.

The company’s Skinsight technology, introduced at CES 2026, uses deep learning to measure skin tightness, hydration, and aging biomarkers from smartphone images. Amorepacific has filed four PCT patent applications around Skinsight and has secured registrations in both Korea and the US.

On the device side, Medicube’s AGE-R Booster Pro (RF + LED + microcurrent multi-modal device) accumulated over 100 million TikTok views after Kylie Jenner was photographed using it at a brand pop-up in Los Angeles. 

APR Corp. (Medicube’s parent company) reached cumulative device sales of 1.5 million units by November 2023, with over 80% of revenue from international markets. The South Korean beauty devices market reached $1.32 billion in 2024 and is projected to grow at a 13.8% CAGR through 2034, reaching $4.81 billion.

What distinguishes the Korean approach:

• Speed-to-market: Korean brands launch hundreds of products monthly; AI-driven consumer feedback loops compress iteration cycles from 18 months to 8–12 weeks.
• Hardware-software integration: Devices are designed to pair with specific serum formulations (e.g., a Medicube device + PDRN serum protocol), creating an ecosystem lock-in similar to Apple’s hardware-software model applied to skincare.
• On-demand manufacturing: SmartSKN’s K-OWN platform and Amorepacific’s COSMECHIP technology enable formulation at the point of sale, reducing inventory obsolescence and enabling a 10,000-SKU-equivalent market with zero overstock.

Amorepacific’s AI Beauty Counselor, developed with Microsoft’s Azure AI stack for Amore Mall, is designed to recommend products based on prior purchases, brand expertise, and ongoing conversation history, making it a strong fit for your report’s “daily beauty tech” and hyper-personalization theme. 

Microsoft’s profile for the launch says the system uses GPT-4o and 4o-mini via Azure OpenAI Service, and that the company plans to add an online skin-diagnosis tool, indicating that Amorepacific is combining conversational AI with diagnostic intelligence rather than offering static recommendations.

The company is also extending this trend into hardware and sensor-led diagnostics. 

At CES 2026, Amorepacific introduced Skinsight, an “electronic skin” platform co-developed with an MIT research group. It analyzes real-time skin-aging signals and delivers personalized care recommendations, while its AI skin analysis technology in Samsung’s AI Beauty Mirror evaluates pores, redness, pigmentation, and wrinkles using a dataset of more than 450,000 cases. 

The same CES announcement links those diagnostics to makeON devices, such as the ONFACE LED Mask and Skin Light Therapy 3S, showing how Amorepacific is building an integrated software-plus-device ecosystem around personalized skincare.

Commercially, Amorepacific is rolling out this model via AMORE CHAT on Amore Mall, where generative AI compares products, summarizes reviews, answers beauty questions, and refines recommendations as users continue interacting.

That makes Amorepacific a strong case study for Trend 5 because it demonstrates how Korean beauty companies are turning AI from a discovery tool into an end-to-end service layer spanning diagnosis, recommendation, and treatment guidance.

Strategic Implications

These trends in this report are not independent signals. They are a systematic convergence of clinical science, biotechnology, and digital intelligence into a redefined standard for skincare efficacy. 

For R&D leaders outside Korea, the question is how to act quickly and in which sequence. 

Three priorities to take center stage.

Build ingredient platforms, not product pipelines. The brands generating a durable competitive advantage in 2026 are not launching products. They are constructing ingredient-led ecosystems that span clinical channels, OTC retail, and consumer devices. Every R&D investment decision should be stress-tested against one question: Does this ingredient anchor a scalable franchise, or does it serve a single formulation cycle?

Compress the clinical-to-commercial translation window. Korea’s structural advantage is not discovery alone; it is velocity. Clinical-grade actives move from in-clinic protocols to mass-retail formats in under 24 months, at price points that democratize access by a factor of six to ten. 

For R&D teams operating outside Korea, the immediate audit is internal. Find which regulatory, procurement, and formulation bottlenecks are extending your translation timeline, and which can be structurally eliminated.

Treat AI personalization as core infrastructure, not a marketing layer. Amorepacific’s diagnostic platform, Samsung’s AI Beauty Mirror, and SmartSKN’s point-of-sale formulation capability signal that the next durable competitive moat in beauty will be a proprietary data-and-device ecosystem. 

Organizations that defer AI skin diagnostics as a future capability risk ceding the personalization layer entirely to platform incumbents.

The strategic window is compressing. The evidence is already in the market. The decision is whether to lead it or follow it.

The post 5 Innovation Trends Leading the Korean Beauty Industry in 2026 appeared first on GreyB.

Energy-related CO₂ emissions reached a record 37.8 gigatons in 2024, while global natural gas demand hit an all-time high and continues to soar with the recent events. This creates a difficult gap for industry: modern economies still need dense hydrocarbon fuels like natural gas and methanol, but producing them from fossil sources continues to increase atmospheric carbon. Direct air capture is gaining attention, but a key commercialization gap remains. Captured CO₂ is still treated as a carbon-management liability rather than as feedstock for valuable products.

Terraform Industries is trying to close that gap by making synthetic natural gas and methanol from sunlight and air, or precisely captured CO₂.

To better understand how they are doing it, we spoke to Casey Handmer, CEO of Terraform Industries. This article contains notable highlights from our entire conversation.

Also Read:

4D Biomaterials Making New Generation of Resorbable Medical Devices – Insights from Phil Smith, CEO

Read Post

How Personalized Medical Device Startups Are Shaping Healthcare in 2025

Read Post

This interview is part of our exclusive Scouted By GreyB series. Here, we speak with the founders of innovative startups to understand how their solutions address critical industry challenges and help ensure compliance with industry and government regulations. (Know more about startups scouted by GreyB!)

“If anyone can produce hydrogen for around a dollar a kilogram and CO2 for around 10 cents a kilogram or a hundred dollars a ton, then they’re in contention economically.”

– Casey Handmer

Dr. Casey Handmer is the founder and CEO of Terraform Industries. He is triple-majored in pure mathematics, applied mathematics, and physics at the University of Sydney and holds a Ph.D. in theoretical physics from Caltech. Handmer was part of NASA’s Jet Propulsion Laboratory, where he worked across GNSS science, Mars mapping, and advanced technology development.

He also worked with Hyperloop One. Handmer left NASA, started Terraform in his garage, and secured $5 million in seed funding to prove his central thesis: that cheap solar electricity had finally made it economical to synthesize hydrocarbons from air and water rather than extract them from the ground.

Having assembled a world-class engineering team and raised $26 million in funding to date, Handmer has positioned Terraform Industries as a potential reinvention of how civilization sources its energy.

Synthetic natural gas and methanol from sunlight and air

Instead of selling carbon credits, Terraform is using captured CO₂ as a feedstock to produce fuels and chemicals such as synthetic natural gas and methanol, targeting markets that already have large-scale demand. Methanol alone is a multi-billion-dollar global market, while natural gas remains one of the world’s largest energy markets.

Its strategy depends heavily on reducing capital costs across the system, especially in hydrogen production and CO₂ capture, so the economics can work without relying only on subsidies.

In March 2024, Terraform completed its end-to-end demonstration, successfully producing fossil carbon-free, pipeline-grade natural gas from sunlight and air. It also hit the landmark cost targets for green hydrogen and direct air capture CO2. (more in the interview)

How is Terraform Industries different from other carbon capture or fuel production startups?

Terraform is making cheap synthetic natural gas and methanol from sunlight and air. The main difference is that we approach direct air capture as a profit center rather than a cost center. For us, captured CO₂ is not just something to bury or offset. It is a precursor for making valuable hydrocarbon fuels.

A lot of companies focus on one part of the process. Some capture carbon and sell credits. Others turn CO₂ into niche materials or chemicals. We are focused on a much larger market: fuels that already power modern civilization. From day one, we understood that the key was keeping capital costs low enough to earn a return and fund expansion.

That is why we have been focused across our three major subsystems: hydrogen production, CO₂ capture, and fuel synthesis. If each part is too expensive, the whole system fails. But if we can engineer the cost down, we can make synthetic fuels that are economically interesting, not just technically impressive.

Why is making synthetic fuels from captured carbon so difficult?

The technology has existed in some form for more than a century. The hard part is not proving that you can make synthetic natural gas. The hard part is making it with positive economic value. There have been projects that achieved their technical goals but still produced gas at prices that would never support modern civilization.

The problem is net energy gain and cost. If the process consumes too much energy, uses too much equipment, or requires too much capital, the final product becomes too expensive. A synthetic gas project can work chemically and still fail economically.

That is why this space is so challenging. Some people underestimate the technical difficulty and have no real chance of success. Others understand the difficulty so well that they never try. We are somewhere in the middle: just bold enough to try, and just technical enough to make progress.

Why did you choose hydrocarbons instead of other CO₂-derived products?

The major challenge here is economic scale. Suppose you find a CO₂-derived product that sells for a high price per carbon atom, which helps unit economics. But if the market is too small, you can spend years building a brilliant process that only serves a tiny part of the economy.

I did not start Terraform to work around the margins. Hydrocarbon fuels are part of a massive global market. Natural gas and methanol are not niche products. They are central to energy, industry, and chemical supply chains.

There are many valid approaches to using captured carbon, and I do not wish ill on other entrepreneurs. But our goal is to solve the core problem at the scale where it matters most. That is why we chose fuels.

What is the biggest cost factor you are trying to solve?

The two key numbers we look at are hydrogen at around $1 per kilogram and CO₂ at around $100 per ton. If anyone can hit those figures, the economics become very interesting. Conventional wisdom says there is still roughly a factor-of-ten gap, so we have been pulling every trick we can think of to bring costs down.

Our intermediate goal was to reach around $2 per kilogram for hydrogen and $200 per ton for CO₂. We are now around that level internally. There is some spreadsheet work involved because we are still running test systems, but we are not pretending labor or real-world costs are zero.

The next step is another factor-of-two reduction. That is harder than the first one, but we are much smarter now. We know where the costs are, and we know which parts of the system need to improve.

What is different about your electrolyzer technology?

Conventional green hydrogen systems are expensive partly because electrolyzers can cost thousands of dollars per kilowatt. If your equipment is that expensive, even perfect uptime and perfect efficiency do not make the economics work.

Our system is under $100 per kilowatt right now. I sometimes compare it to the cost of a very cheap microwave. My goal is to bring it closer to electric kettle territory, around $20 to $30 per kilowatt.

We achieved this through repeated engineering iteration. For example, our previous prototype used a temporary construction method that simplified manufacturing but added cost. We later moved to a plastic welding process, which helped remove around half of the remaining electrolyzer cost. These are not flashy breakthroughs. They are hard engineering improvements that compound over time.

How low is your hydrogen cost today?

Our best internal estimate is around $2 per kilogram, probably slightly under that. This is based on our current test systems and our understanding of how the productized system will perform.

That number matters because conventional green hydrogen is often discussed at much higher prices. From what we see inside the industry, electrolysis-based green hydrogen is rarely anywhere close to where it needs to be for synthetic fuels.

We are already showing that the existing model is weaker than many people assume. Our internal cost basis is below $2 without depending on large subsidies. That gives us confidence that our approach can work.

How does Terraform’s cost strategy support expansion?

Expansion only works if the economics work. If your first plant is too expensive, you cannot use it to fund the next one. That is why we have focused on capital cost from the beginning.

Our view is that low CapEx gives us a higher return on investment, which then allows us to expand the technology. This is not just about building one impressive machine. It is about building a system that can reproduce itself economically.

That is also why we focus on large markets. If the product is valuable, the market is enormous, and the system can be built cheaply enough, then growth becomes much more practical. The whole company is designed around that idea.

Meet our Interviewer – Shabaz Khan, Marketing Manager at GreyB

Shabaz Khan, Marketing Manager

Shabaz, is a seasoned marketing manager and leads the Scouted By GreyB. With a decade of experience, he specializes in delivering critical insights to Innovation leaders, R&D, and IP teams about evolving tech landscapes, innovation trends, and emerging breakthrough startups. Shabaz excels at aligning research data with business needs and developing strategies to solve innovation challenges. His leadership and problem-solving skills make him a valuable asset in R&D and IP research.

Want to find other scalable startups working on carbon capture and its utilization? Please fill out the form below to contact our experts. 

The post This Startup Generates Fuel From Air, Water, and Sunlight appeared first on GreyB.

You run the AI search. The tool returns 40 references in under two minutes. The top results look relevant. The technology domain aligns. The concept matches. You feel like the search is heading somewhere.

This is precisely when the risk starts.

AI tools have changed the pace of patent searching. Semantic search finds conceptually similar documents faster than any Boolean string. Re-ranking engines process thousands of results in seconds. The coverage feels comprehensive. But coverage and completeness are not the same thing. In prior art searching, completeness is the only metric that matters when litigation is the downstream consequence.

The industry already recognizes this. As Lexology noted in early 2026, AI improves patent search but doesn’t eliminate the need for expert judgment because legal relevance stays contextual, depending on claim construction, jurisdictional standards, and procedural posture. PatSnap’s research team put it more directly: general-purpose AI struggles with the precise legal interpretation of novelty and non-obviousness as applied in patent law.

The practical consequence is that researchers who rely on AI-ranked results without interrogating what those results represent can walk away from a search believing they’ve found the relevant prior art, when they’ve found only what the AI was trained to surface.

We covered this in depth in a recent GreyB webinar on AI in patent searching, where our researchers walked through the specific failure points they encounter in live projects. Watch the recording here. 

The First Trap: Over-Reliance Looks Like Efficiency

Many researchers start using AI with one expectation: if a relevant reference exists, the system will find it. That expectation is the trap.

An AI tool that fails to surface strong prior art doesn’t signal its own failure. The analyst sees a list of results, reviews them, and draws a conclusion. If no strong reference appears, the assumption becomes that no strong reference exists. The prior art may be hidden behind different terminology, a classification the tool didn’t reach, or non-patent literature the model deprioritized. None of that is visible in the output.

As Divyansh, Senior Research Analyst at GreyB, framed it:

“The biggest risk is not just missing the prior art. It is becoming too confident in the completeness of the search. AI identifies patents based on semantic similarity, but it doesn’t inherently perform claim interpretation or reason through the claim scope the way an experienced analyst would.”

This is also where AI outputs produce what practitioners describe as false confidence. A large language model asked to find prior art for a claim will present loosely related references with the same certainty it presents strong ones. This happens partly because prior art searching is a specialized task. These models are trained on general language patterns. They’re built to find papers related to a topic, not to test a document against a specific set of claim features. Topical similarity becomes the proxy for claim relevance. The two are not the same.

Semantic Similarity Is Not Claim Relevance

AI retrieval works through vector similarity. It finds documents where the language and concept profile resemble the input. That’s useful for narrowing a field. It’s structurally different from how a patent professional evaluates a reference.

A researcher working on a claim on prediction models needs references that disclose that specific model type. AI surfaces references to classification and recommendation models because they sit in the same machine learning domain and share semantic features. The documents are related. They don’t map the claim.

Mahesh Maan, Technical Architect at GreyB, separates this into two problems that most tools conflate. Pulling relevant documents from a database is one task. Analyzing whether a document actually discloses the claim features is another. Generative AI handles the second reasonably well.

The first handles through vector search, which retrieves documents that are textually similar but not necessarily precise at the claim level.

GreyB’s published case study on how technical context cracked a patent invalidity case illustrates this directly. The inventive element in that case was extracted from the prosecution history, not from the claim text itself, and the search that found the decisive prior art used a three-word query, not a semantic description of the full claim.

A researcher can complete the search, review 40 high-confidence results, and still not have covered the actual prior art landscape. The tool produced results. The search isn’t done.

Five Claim-Level Traps That Show Up Consistently

The gap between semantic similarity and claim relevance shows up in five recurring patterns. Each one produces a plausible-looking result that fails on closer inspection.

Fine-Grained Limitations

A reference can match the top-level concept of a claim and miss a single sublimitation that determines whether the reference is strong. A vehicle monitoring claim might disclose the system architecture clearly. If it doesn’t trigger an alert after a specific threshold duration, that sub-feature is absent. The reference looks strong at first glance. It isn’t.

AI systems are good at identifying broad concept matches. They’re weaker at detecting the fine-grained claim limitations that define actual scope. The unconventional invalidity tactics that turn difficult cases often come from a close reading of prosecution history to isolate exactly which sub-feature the patentee added to survive examination, then targeting that specific element rather than the broad concept.

Conditional Triggers

Claims that include “when” logic or “in response to” language require more than presence. They require causality. AI will often map “in response to” language to a reference where step B follows step A, because the sequence is there. The reference may not establish that A causes B. In litigation, that distinction is the difference between a usable reference and one that gets challenged.

Kush, Team Lead at GreyB, made the point directly:

“Sometimes the invention is not about what it does. It’s about when and why. If the condition isn’t present in the reference, the reference is never useful, regardless of how well the rest of it maps.”

AI systems consistently overlook this granular detail. They produce a first-level mapping that looks complete. Human review has to verify the causality, not just the sequence.

Negative Limitations

What a claim excludes is as important as what it requires. AI systems optimized for semantic similarity bias toward documents that discuss the same components. A claim requiring a single-wire interface, in which the absence of a second wire is the inventive step, will yield results full of multi-wire systems because those documents use the relevant terminology most frequently.

Searching for what’s absent is genuinely hard. Even experienced manual searchers build specific strategies around terms like “without,” “switchless,” “contactless,” or anything ending in “less” to capture these patterns. AI search, especially vector-based retrieval, works in the opposite direction. Asking it to find patents that don’t transmit BPDUs will bias the vector toward patents that do.

Mahesh Maan describes the gap this way:

“Identifying these results from the database is a very big challenge for today’s AI. If you tell a vector-based system you want patents without transmitting BPDUs, the vector will have a strong affinity toward patents that are actually transmitting BPDUs. It captures the opposite of what you need. When it comes to analysing a result that’s already in front of it, AI can work. But finding that result from the database in the first place — that’s where it fails.”

The practical lesson is that a researcher has to slow down and ask explicitly: what does the claim exclude, what must not be present, and what conditions must be absent. These questions can’t be left to the tool.

Sequence and Order

When a claim requires steps in a specific sequence, say detection, then authentication, then access, a reference that discloses the same steps in a different order can appear relevant but not anticipate. AI tends to assess component overlap without checking order.

This matters in FTO searches as much as it does in invalidity. A product that performs the same operations in a different sequence may not infringe. A reference that performs the same steps in a different order may not anticipate. AI often treats both as matches.

With longer sequences, eight or nine steps, this becomes more severe. The AI loses track of the required flow and often defaults to the more common sequence it has seen in training data, even when the claim specifies something different. The error repeats across all references it analyses because it hasn’t correctly parsed the claim in the first place.

Element Dependencies

A reference that discloses a sensor and a control unit can rank highly against a claim that requires the control unit to act on signals from the sensor. The components match. The functional relationship may not. AI flags the components as present without establishing whether the required interaction between them exists.

This connects to a deeper structural problem. Claims carry hierarchy through indentation in a patent PDF. When AI receives claim text from a database, that visual structure is often absent. The AI sees a flat list of features and analyses them independently. It may confirm that a reference has a memory register without recognizing that the memory register is only meaningful if the auxiliary processor it belongs to also exists. If the auxiliary processor isn’t present, the memory register analysis is irrelevant.

The fix is explicit. Tell the AI the relationship between elements, not just the elements themselves. Prompt it to track dependencies across limitations, not just feature presence.

Terminology Changes. Most AI Synonym Lists Don’t Reflect That.

Patent terminology evolves across generations of technology. Cloud computing was described as distributed network processing or remote server-based computing fifteen to twenty years ago. A machine learning claim has prior art in papers discussing statistical pattern recognition and adaptive algorithms. The concept is the same. The language is not.

AI handles well-known synonyms well. Ask a language model for synonyms of “mobile phone,” and it produces a reasonable list. But it defaults to generic terms in the first pass. It won’t volunteer that UE was the dominant terminology in 3GPP LTE filings while WTRU was standard in earlier CDMA specifications, unless you prompt it specifically from that angle.

The AI has this knowledge. It doesn’t surface automatically. Mahesh Maan describes this as eliciting latent knowledge: the model holds it, but you have to pull it out from specific angles. Ask what terms were used in a specific decade. Ask how a particular standards body described the concept. Ask how it would have been framed in the context of a specific platform or regional filing practice. Each prompt pulls different terminology. The combination covers more ground than a single synonym query.

GreyB’s ten patent search tips from expert researchers cover this in detail, including how to use CPC class definitions to extract terminology that the technical literature uses in a specific domain, which is often different from what a language model produces in a generic synonym pass.

A researcher who accepts the first synonym list and builds a keyword strategy around it is leaving prior art in the database.

Classification and Citation Networks Are Not Optional

Text-based AI search ignores two of the most reliable prior art discovery paths: CPC classification and citation trails.

Classification organizes patents by technical concept, not by language. A patent on video streaming optimization filed in the mid-2000s may not use any modern terminology around adaptive bitrate streaming. It sits in a CPC class related to bandwidth management. Semantic search against a modern claim description won’t find it. A search starting with the right classification code will. GreyB’s Google Patents search guide explains how to apply CPC codes in practice, including how to use the classification hierarchy to move from broad to narrow and avoid pulling irrelevant results.

Language models know the CPC system. They can suggest relevant classes for most technical concepts, and they do so accurately when asked. They don’t use that capability by default when building search strings. They produce keyword queries. If you don’t explicitly prompt the AI to incorporate classification codes, it won’t.

Mahesh Maan puts the stakes clearly:

“Classification is one of the most powerful tools for patent searching. If an AI is not leveraging it, it’s leaving a lot on the table. But the burden of using classification falls on the user. The LLM already has this capability, but it won’t use it on its own.”

The same logic applies to using a found reference as a starting point. Feed a useful result back to the AI and ask it to identify CPC codes or terminology patterns from that reference that could expand the search. This feedback loop often surfaces classes or terms that the initial strategy missed.

Citation networks present a different scale problem. Ten citations, each with ten citations, create a tree that exceeds what a standard AI context window can efficiently manage. Manual searchers develop a sense for which citations to follow based on titles alone. Experienced researchers can scan 50 citation titles and know which two are worth pulling. Tools like Ambercite use citation-based AI ranking to make this more systematic, but even there, the analyst directs which part of the citation tree to enter. AI doesn’t match that pattern recognition reliably when left to work autonomously. It processes citations sequentially and loses context across a large tree. 

Non-English Filings Create a Structural Blind Spot

A significant portion of global prior art sits in Japanese, Korean, and Chinese filings. For technology areas with deep Asian development histories, battery thermal management, semiconductor interfaces, and telecommunications protocols, the strongest prior art may live in regional databases in documents never indexed with English-first terminology.

Machine translation has improved substantially. But older filings, translated before large language model tools were available, carry errors that distort both keyword search and semantic retrieval. A translation artifact that renders “Venetian blinds” as “dried bean curd” is not an edge case. It’s an example of how translation quality directly affects whether a document ever gets retrieved.

As Divyansh noted:

“The strongest prior art doesn’t always appear in an English-language search. Highly relevant references may exist in older Japanese, Korean, or Chinese filings that use completely different technical phrasing or region-specific drafting styles. Even when translations are available, automated systems may fail to preserve technical nuance. This creates a major blind spot. The search may appear comprehensive while critical prior art remains hidden.”

GreyB’s guide to advanced prior art search strategies covers the geography-first approach in detail, including why certain technical domains map to specific national patent offices and how to target JPO or KIPO when the technology’s origins point there. Separately, GreyB’s breakdown of advanced patent databases for invalidity searches lists the region-specific databases that are most useful when standard global platforms return limited results.

AI-powered semantic search of translated documents inherits the quality of the translation. A well-translated document is retrieved accurately. A poorly translated one may not retrieve at all, regardless of how relevant the underlying content is. Knowing which technical domains have significant non-English prior art and checking translation quality before drawing conclusions about completeness are part of the search process, not optional steps.

The First AI-Ranked Result Is Rarely the Best Disclosure

The top result from an AI-ranked search is not always the strongest version of a reference. US filings tend toward narrower descriptions. A Japanese priority application may contain broader figures, clearer implementation diagrams, or technical details that were narrowed during prosecution in other jurisdictions. A system architecture claim may appear in one family member. The specific workflow that maps the claim limitation appears in another.

AI tools don’t automatically check patent families when analyzing results. A reference from a jurisdiction with a post-priority date may be less useful than a family member filed earlier, and that family member may disclose the missing limitation clearly. An experienced searcher checks this instinctively. An AI agent does it only if explicitly instructed to do so.

Divyansh framed the implication:

“Relying only on the primary AI result can cause researchers to miss stronger disclosures, earlier priority support, or more complete technical explanations hidden within the patent family. A US filing may contain a narrower description while its earlier Japanese priority application includes broader figures or disclosures that were later removed during prosecution.”

Any AI-assisted search workflow needs explicit rules around patent family handling. Otherwise, the most useful version of a reference never reaches the analyst’s desk.

The Patent Search Framework Has to Evolve With the Tools

Before AI, patent searching had a defined process. A set of documented best practices, followed in sequence, produced a defensible search. AI has made that process harder to define, not easier.

More can be done now. An analyst can submit 3,000 results from a single CPC class to a re-ranking agent and find the strongest reference in the top 30. That wasn’t operationally feasible before. The same analyst can use AI to reconstruct the competitive and technical context around a patent’s priority date in minutes rather than days. The current landscape of AI-based patent search databases reflects this: tools now exist specifically for re-ranking, citation mapping, semantic concept search, and non-patent literature retrieval, each suited to a different part of the search problem.

But more options also mean more ways to spend time poorly. AI tools can absorb an entire working day on a technology where a well-built Boolean string would have produced better results in an hour. Some technologies yield strong results through AI retrieval. Others require manual keyword construction, and applying the wrong approach wastes time and produces worse outcomes.

Mahesh Maan describes the required shift:

“It’s no longer a set process. You have to get creative, understand where the AI lacks, and be adaptive in terms of what is the best thing you can do at each stage. If you let AI take the driver’s seat, it will always suggest what it can search next. But if you stay in the driver’s seat, you use AI and traditional tools to get where you need to go. The AI can do a lot of the leg work. What you need to do is maintain a clear view of where you are and where you need to go next.”

The most effective approach uses AI to shortlist, re-rank, and generate terminology, while keeping classification, citation analysis, negative limitation searches, and family checks as human-directed steps. Neither AI alone nor traditional methods alone produces the best outcome. The combination, with a researcher directing the process, does.

Before Calling a Patent Search Complete, Answer These Five Questions

1. Have the AI results been mapped against specific claim limitations, not just the general concept?
2. Have the conditional triggers and “in response to” language been verified for causality, not just sequence?
3. Has the search covered what the claim explicitly excludes, with a strategy built around finding references where that feature is absent?
4. Have non-English filings been reviewed, and has translation quality been checked for the relevant technical domain?
5. Has CPC classification been applied, have citation trails been followed beyond first-page results, and have patent family members been reviewed for stronger disclosures?

If any answer is uncertain, the search has gaps. In a high-stakes litigation context, gaps are where the other side builds its case.

The prior art that invalidates a patent or defeats an FTO clearance doesn’t always sit in the most obvious place. Identifying where your current search strategy falls short before a court does is the difference between a defensible position and an exposed one.

If you’re running invalidity searches, FTO analyses, or validity studies where search completeness is at stake, identify the gaps in your current search approach.

The post 2 Minutes of AI Patent Search & 1 Missed Non-English Filing Will Cost $4M in Litigation appeared first on GreyB.

Most water monitoring facilities still rely on weekly or biweekly lab sampling. This method cannot detect a harmful algal bloom until it has already reached intake points. The US EPA identifies algal toxin removal as one of the most cost-escalating operational challenges for drinking water treatment. Studies of affected utilities document measurable spikes in labor, chemical spending, and cartridge-filtration costs during bloom events.

Aquaculture faces a similar problem to that of water treatment facilities: long testing timelines and delayed results. Disease outbreaks can spread faster than lab testing can be completed. Globally, disease-related aquaculture losses have been estimated to exceed $6 billion annually, with parasite-related losses alone estimated between about $1.05 billion and $9.58 billion.

NS2 or Nucleic Sensing Systems solves this by bringing droplet digital PCR-style testing out of the lab and into the field. Instead of taking a water sample, shipping it to a lab, and waiting days or weeks, NS2 delivers quantitative results within about an hour.

To better understand how they are doing it, we spoke to Ed Rudberg, CEO of Nucleic Sensing Systems. This article contains notable highlights from our entire conversation.

This interview is part of our exclusive Scouted By GreyB series. Here, we speak with the founders of innovative startups to understand how their solutions address critical industry challenges and help ensure compliance with industry and government regulations. (Know more about startups scouted by GreyB!)

Also Read:

The solution to toxic algae was hiding in a steel mill and this startup found it!

Read Post

This startup fetches 99% of valuable lithium from polluted wastewater!

Read Post

“There is nothing better than having good, smart people around you.”

– Ed Rudberg

Dr. Edgar (Ed) Rudberg has brought multiple companies from product ideation to profitability. He holds a Ph.D. in Natural Resources Science and Management and a Master’s in Marine Affairs and Policy. He also serves as CEO of CD³, a General Benefit Corporation and NS²’s parent company, which manufactures decontamination infrastructure for boat ramps. This role organically led him toward the biological sensing challenge NS² was built to solve.

Under his leadership, NS² earned the Midwest regional win at the Cleantech Open and reached the national finals. Rudberg has been instrumental in securing SBIR grant funding for NS² and making the university’s research technology a market-ready platform.

NS2 turns water into real-time biological data

NS2, or Nucleic Sensing Systems, helps drinking water utilities and aquaculture farms detect and quantify organisms in water without waiting for a traditional lab workflow. Instead of treating water testing as a slow, periodic process, the company wants to make it continuous, automated, and fast enough to support operational decisions.

The system takes inspiration from droplet digital PCR, a gold-standard human diagnostic technology. NS2 has adapted that capability into a box that can be deployed outside the lab, removing the need for a trained human operator at every step. Its flagship product, the Tracker, has achieved field deployments in federal hatcheries and secured government sales.

What technology is NS2 working on, and what are you trying to bring to market?

At NS2, we’re building a real-time biological transmission system that turns water into continuously monitored data streams. In many traditional systems, whether it is aquaculture or drinking water, customers take a water sample, send it to a lab, and wait for analysis. That process can create a delay of days, weeks, or sometimes even months.

What we are doing is giving customers sample-to-answer data on the biology in their water within about an hour. That means they can act before a problem becomes expensive or dangerous. Instead of reacting after a disease outbreak, biofouling event, or public health concern has already developed, they can make proactive decisions.

Which industries are you targeting first?

We cannot partner with every industry at once, so we are starting with drinking water utilities and aquaculture. In drinking water, algae can create sludge and other operational problems that can cost even a mid-sized utility more than a million dollars a year to manage. Early detection gives utilities a better chance to respond before the issue gets out of control.

Aquaculture is another strong early market because disease losses are a major industry-wide problem. Fish farms can lose large amounts of money when disease moves through a system quickly. We see a real opportunity to help these operators detect changes earlier and reduce losses.

How does your system bring testing time down from days to hours?

Think about the PCR tests many of us used during COVID. A sample was taken, sent out, and analyzed to detect whether a specific virus was present. We are using a similar idea, but instead of swabbing a person’s nose, we are essentially swabbing the environment.

The technology we use is based on droplet digital PCR. It breaks a sample into tens of thousands of tiny sub-samples. Each one gives us a signal, almost like a one or a zero. By doing this, we digitize biology and can quantify how much of a target organism is present over time.

That matters because many organisms are not simply present or absent. In aquaculture, for example, a disease organism may always exist in the system at some low level. The important question is whether its concentration is rising. Our system helps customers see that change early and then track whether their mitigation efforts are working.

Can the system detect anything in the water automatically, or does the customer need to know what they are looking for?

The customer does need to know what they want to look for. We program the machine for the organisms or biological targets that matter to that customer. It is not a system that flags every unknown organism automatically.

That said, our system can look for different organisms at different times. We run in a linear fashion, almost like planes lining up on a runway. At one point, we might look for five organisms, and 15 minutes later, we can look for five different ones. So while the system must be programmed, it still gives customers the flexibility to monitor multiple targets over time.

How did you design the system for reliability and long-term maintenance?

We built the system like LEGOs. It is made up of a series of subsystems that work together and talk to each other to create a functioning machine. That design choice was important because PCR machines can last for years, but field systems need to handle real-world wear and tear.

If something fails, we do not want the customer to replace the entire machine. We want them to replace a subsystem. That makes maintenance easier and more practical.

It also made development more manageable for us. Instead of trying to perfect one giant system all at once, we could focus on each subsystem and make it stronger. That approach helps us build a more robust product for customers.

How are you proving accuracy compared with traditional lab-based testing?

We have been working closely with the US federal government, which has been very forward-looking in environmental DNA detection. Every organism sheds DNA into its environment, whether it is a dolphin, a fish, bacteria, or something else. We are using that idea to monitor biological signals in water.

One important point is that what we detect in the field is not always exactly the same as what traditional lab methods detect. Many standard methods collect cellular environmental DNA by filtering water and sending the filter to a lab. But cells and free DNA behave differently in the environment. DNA does not stay there forever; it degrades.

That is why comparison work is so important. Our partners have taken water samples, flash-frozen them, brought them back to the lab, and analyzed them so we can compare their results with ours. That kind of side-by-side validation helps show the efficacy of our technology.

How do you address cost compared with traditional monitoring?

We constantly look at our bill of materials and ask what the most expensive components are and how we can reduce those costs. Optics are a good example. At one point, we were looking at very expensive laser systems. As LEDs and lasers changed in price and performance, we kept re-evaluating what could work best.

Traditional lab testing might cost around $115 to $175 per sample, depending on what is being tested. But the bigger problem is the time lag. If an aquaculture disease can wipe out a system before lab results come back, then the old process does not meet the customer’s real needs.

Our value is in giving a sample-to-answer result within about an hour. Over time, this can support automated systems, machine learning, reduced chemical use, lower energy costs, fewer antibiotics, and better operations. One customer looked at our technology and believed it could create a 46% return for a drinking water utility when considering the full monitoring workflow and human labor involved.

Are there regulatory barriers for drinking water or aquaculture adoption?

We are intentionally avoiding heavy regulation at this stage by focusing first on operational use cases if we needed federal approval before customers could adopt the technology, that would create a long and costly delay between launch and market adoption.

In aquaculture, the data is mainly used by operators to reduce disease losses. It is not being used to certify that the food is safe for humans. In drinking water, we are initially focused on things like harmful algae from an operational standpoint, not as a formal toxicity compliance test.

That means we can help with issues such as biofouling, taste, odor, and treatment operations. People expect the water coming out of the tap to taste and smell acceptable. Our technology can help utilities manage those quality and operational issues earlier.

Meet our Interviewer – Shabaz Khan, Marketing Manager at GreyB

Shabaz Khan, Marketing Manager

Shabaz, is a seasoned marketing manager and leads the Scouted By GreyB. With a decade of experience, he specializes in delivering critical insights to Innovation leaders, R&D, and IP teams about evolving tech landscapes, innovation trends, and emerging breakthrough startups. Shabaz excels at aligning research data with business needs and developing strategies to solve innovation challenges. His leadership and problem-solving skills make him a valuable asset in R&D and IP research.

Want to find other scalable startups working on sustainable water treatment solutions? Please fill out the form below to contact our experts.

The post This Startup Brings Lab-grade DNA testing to Drinking Water Treatment Plants appeared first on GreyB.

Of the 800-plus SEP invalidation searches GreyB completed in the past year, the decisive prior art came from outside patent databases in most cases.

The key evidence often did not come from a granted patent or a published research paper. It came from a 3GPP working group proposal, a draft contribution, an IEEE meeting document, or an email thread written before the final standard was published.

That matters because most search workflows are not built to handle this type of material. Standard document databases and SSO archives are built for document retrieval. Boolean search is constrained, cross-source filtering is absent, and AI-assisted retrieval is unavailable in most SSO systems. Analysts often open documents one by one, follow trails that lead nowhere, and work through tens of thousands of pages while facing litigation holds, IPR filing deadlines, or FRAND negotiations.

The results from those 800-plus searches show a clear pattern. In 70% of cases, the search produced strong results. More than 60% of those searches produced Category X, 1, or 2 references. These references can directly challenge novelty and support strong invalidation arguments. The remaining 40% produced Category Y references and combinations that supported obviousness arguments.

Those results did not come from search expertise alone. They came from six AI tools operating alongside analyst judgment, each used to remove friction at a specific stage of SEP prior art search where manual-only workflows lose time, miss sources, or stop short of the most consequential evidence.

What follows is a description of each tool, what it does, and where it changed the outcome of a real SEP patent invalidation project.

The Six-Layer Search Architecture Behind GreyB’s SEP Invalidity Work

SEP invalidity searches fail when the standard is treated as a document repository rather than a development record. The analyst first needs to reduce the asserted claim to the exact technical feature it covers. That feature then needs to be mapped to the standard, traced through older drafts and meeting contributions, and tested against the patent’s priority date.

GreyB’s workflow follows that sequence.

The process combines analyst judgment with six internal systems that reduce the most failure-prone parts of the SEP prior art search. These systems support claim interpretation, essentiality confirmation, standard backtracking, meeting document retrieval, email archive search, technology evolution mapping, and clustered patent search.

Each layer narrows the search field and opens more precise search paths.

The tools do not replace SEP search expertise. They remove friction from the work that slows analysts down, including reading dense claims, locating standard sections, searching thousands of proposals, tracing working group discussions, and building a timeline of how the claimed feature evolved.

Also Read:

AI Just Made SEP Claim Mapping 80% Faster – 600 Hours of Human Work to 14 Days

Read Post

The analyst still decides which references matter, how they map to claim elements, and whether they support anticipation, obviousness, or a stronger invalidity position.

Strong SEP prior art rarely comes from a single keyword query. It usually appears through a sequence of steps. The analyst needs to understand the claim, confirm overlap with the standard, identify the technical feature, trace the feature backward, locate the earliest public disclosure, and test that disclosure against the priority date.

1. The Hour Lost Before Search Begins and How to Reclaim It

The first goal in any SEP prior art search is building a precise understanding of the claim. This sounds straightforward, but it is where analysts lose the most time and where many interpretation errors begin.

SEP claims are technically dense. The underlying technology is often complex, and the claim language is compressed. The entities, conditions, and interactions are difficult to visualize from text alone. An analyst working only from written claim language starts at a disadvantage.

The Advanced Patent Understanding Tool addresses this directly. Input the patent number, and the tool generates a structured breakdown of the invention, the problem it addresses, the proposed solution, and the narrowest defensible reading of each claim element. Alongside this, the tool renders the claim as a visual diagram, making communication flows, entities, and system conditions immediately readable.

In one invalidation project, the claim described communication between a UE and a base station through a relay node with specific CSI-RS transmission conditions.

An analyst would normally need close to an hour to fully understand the claim from the text alone. The tool reduced that work to minutes. The diagram surfaced the system architecture, the relay’s role, and the exact conditions governing CSI-RS transmission. Those details shaped the later search strategy.

The tool also detects issues that text analysis alone may miss.

In another project, the tool flagged a contradiction in the phrase “wherein the non-essential minimum SI is a minimum SI.”

The term “minimum SI” refers to information required for initial cell selection and camping. “Non-essential SI” is not required during that process. The AI agent identified this as a specification failure to clearly define “minimum SI” in context, stronger than a drafting irregularity, and sufficient to open an indefiniteness position that was folded into the combined technical and legal strategy.

This foundational step produces the precision of claim understanding on which every subsequent search decision depends.

2. Most Declared SEPs Don’t Truly Map, Finding Out Which Ones Do Changes the Search Path Entirely

Before investing significant effort in a prior art search, the threshold question is whether the patent genuinely reads on the relevant standard, whether it has a verified claim-to-standard mapping rather than a broad declaration of essentiality. The gap between declared essentiality and verified essentiality is wide. Proceeding with a full prior art search on a patent that does not survive essentiality verification is an avoidable waste of search resources.

Verifying essentiality manually means working through hundreds or thousands of pages of standard documentation to locate sections that overlap with the claim. For an experienced SEP analyst, that process takes five to six hours per claim. The risk of missing a relevant section is real, particularly when the relevant language appears in a different section from where it would be expected.

The AI SEP Mapping Tool compresses that process to minutes. Input the patent number, the claim, and the relevant standard. The tool surfaces a mapping table identifying candidate sections for overlap. The analyst then manually verifies those sections, the tool identifies where to look, and the analyst confirms what is there. 

When genuine overlap is confirmed, the tool’s value extends further. The standard can be backtracked to before the priority date, allowing identification of which company first introduced the relevant concept into the standard development process. That company’s patent portfolio in the same period then becomes a targeted prior art search in its own right. The path from confirmed essentiality to originating contributor to their patent filings is one of the most reliable routes to Category X prior art. 

The tool identifies the place where prior art is most likely to exist.

3. The Prior Art That Lives in Meeting Rooms, Not Patent Databases

For SEP invalidations, the most valuable prior art often sits in meeting proposals, technical contributions, and working group documents produced during the years-long process of building a standard. Finding it requires a full-text search across a body of material that can run to tens of thousands of documents.

The Standard Proposal and Meeting Document Search Interface makes that tractable. Boolean queries run across the full text of documents from 3GPP, IEEE, and video codec bodies, including JVT, JCTVC, and JVET. Filters for specific meetings, working groups, companies, and date ranges allow the analyst to narrow the search with precision. What would otherwise require manually opening and reading document after document becomes a structured, reproducible search.

An AI layer sits on top of the manual interface. It constructs its own search queries, runs iterative searches across selected working groups, and returns a ranked list of results, surfacing the most directly relevant leads early. This is particularly valuable for quickly identifying obvious prior art and ensuring that well-documented references are not missed due to an imperfect initial query formulation.

The two interfaces work as complements. The AI casts the first net broadly and efficiently. The analyst then refines and extends the search with targeted queries based on the early results. 

4. The Prior Art That Predates the Proposal and Where to Find It When 3GPP Returns Nothing

A recurring situation in SEP invalidation work is this: the most directly relevant meeting proposal in a standard development thread appears just after the priority date, days or weeks too late to be usable as prior art. But the ideas in that proposal did not originate at the meeting at which it was filed. Working group delegates regularly discuss and develop concepts in email threads before those ideas formally appear in a proposal document. Those email discussions, when they predate the priority date, are prior art.

The challenge is scale. Email archives are vast. Searching them manually – reading through threads, guessing at relevance from subject lines, following trails that lead nowhere – is prohibitively time-consuming. There is also no guarantee that a manual search covers the relevant window.

The Email Archive Search Tool applies Boolean query logic to these archives with date-range filters focused on the window before the priority date. Results are returned with subject lines and content previews, and links go directly to the original archive pages where attachments – including draft versions of proposals – can be checked.

The tool’s value extends beyond email threads to draft standard documents that are no longer publicly accessible. TS 36.323 Version 1.1.1, for instance, is no longer available on the 3GPP website. Yet early draft versions of a standard can be essential to establishing that a particular feature was already described in the art before the priority date.

In one engagement, the tool identified not only the individual who introduced the feature into the standard but also the assignee who worked on it, along with the relevant standard document that had been circulated as an email attachment. This document was not found in any searchable public database.

A search direction that would previously have meant hours of manual archive work can be executed in minutes, with confidence that relevant threads are not being missed.

5. When References Alone Don’t Win: Building the Chronological Record That Settles Cases

Strong prior art references are the foundation of an invalidity case. In litigation and IPR proceedings, they are not sufficient on their own. The record also needs to show that the claimed invention did not emerge from a moment of inventive insight, that it was a continuous, observable progression of development within the standard, often with contributions from the patent owner itself.

Building that narrative manually is one of the most labor-intensive parts of SEP invalidation work. It requires identifying the sequence of proposals, tracking which features appeared when and in whose contributions, and constructing a coherent chronological account of how the technology evolved up to and through the priority date. Done entirely by hand, this step can consume days.

The Technology Evolution Narrative Generator accelerates it. Feed it the relevant meeting document links, from 3GPP, video codec bodies, or IEEE, and it produces a chronological mapping of which documents introduced which features, and when.

The output gives the analyst a structured timeline and a set of concrete search leads: the companies whose contributions sit closest to the claimed invention, whose patent portfolios then become a targeted prior art search in their own right.

In one project, a client came in during an active SEP dispute over a 5G NR scheduling feature. They had promising prior art candidates but no coherent story connecting them. Opposing counsel had built a strong narrative around the inventor’s contribution to the relevant working group.

Also Read:

5 Powerful techniques to successfully invalidate telecom patents

Read Post

Running the relevant 3GPP RAN1 meeting documents through the narrative generator returned a timeline showing that the core scheduling mechanism had been incrementally shaped across four consecutive meetings, with contributions from three separate companies predating the priority date by over a year. One of those contributing companies held a patent covering an earlier iteration of the same feature, a reference that would have taken days to surface manually.

The analyst used the generated timeline as a structural backbone, refined the sequencing, added technical commentary, and folded the newly identified patent into the written argument. The narrative reframed the entire dispute: rather than a single inventive contribution, the record showed a feature built collectively in the open, step by step, before the priority date. 

The case settled shortly after the invalidity contentions were filed.

6. Five Days, Active Litigation, No Room for a Conventional Search

In a conventional prior art search, analysts build keyword and classification queries and work through large, undifferentiated result sets. That approach is reliable but slow, and it carries the risk of the most relevant patents being buried in noise before the search window closes.

In one matter, the search had to be completed in approximately five days. The domain was Wi-Fi. The client’s patent was already in active litigation, and an injunction was a real possibility without a strong invalidity position demonstrated within the available time.

Also Read:

How we Invalidated a Wireless Communication Patent using Standards?

Read Post

The Technology Cluster Patent Database provided a different entry point. Rather than starting from a blank query across the full patent corpus, the relevant technical node was identified within the cluster database, a pre-curated set of patents organized around specific technology areas, including Wi-Fi 6 features. Applying the search cutoff date and running a targeted query within that cluster returned the highest-probability prior art candidates without the volume and noise of a conventional search.

For Wi-Fi 6 patents in particular, the strongest prior art is frequently found within the cluster without needing to extend the search further. When the cluster falls short, it still provides a high-quality starting point that makes the broader search significantly more efficient.

The SEP Invalidity Search That Finds What Exists, Not Just What Is Easy to Find

A search can return strong references and still miss the working group email that predates the key proposal. An essentiality mapping can be clean and still stop short of the originating contributor’s patent portfolio. The absence of a result reflects the methods used, not the completeness of the prior art landscape.

The six tools described here, the Advanced Patent Understanding Tool, the AI-Based SEP Essentiality Checker, the Standard Document Search Interface, the Email Archive Search Tool, the Technology Evolution Narrative Generator, and the Technology Cluster Patent Database, do not automate SEP invalidity analysis. Analyst judgment still determines which references matter, how they map to claim elements, and whether they support anticipation, obviousness, or a combined invalidity position.

The tools make that judgment more effective. The strongest invalidity case is usually built before litigation positions harden.

The working group documents and email threads that carry the most weight are rarely indexed in patent databases. Finding them requires both the tools to reach them and the expertise to interpret them correctly.

If an SEP invalidity search is already in progress, a deadline is approaching, or the strength of an existing prior art position is uncertain, GreyB’s SEP team is available to discuss the details.

The post 6 AI Tools, One Sequence: GreyB’s Six-Layer SEP Patent Invalidation Process appeared first on GreyB.

An analysis of approximately 3224 US litigation cases from 2021 to 2025 involved ~3922 unique US patents. A deeper dive into those patents revealed that base station-related patents account for 52% of all telecom litigation on the docket. The five most frequently named defendants across that period are AT&T, Samsung, Verizon, T-Mobile USA, and Ericsson.

Base station patent litigation is not a growing threat. 

It has been the dominant reality in telecom IP for five consecutive years. The companies treating it as an emerging risk to monitor have already missed the first wave.

In 2024, Ericsson was named directly in a US antitrust case by Celerity IP, a non-practicing entity that had accumulated base station patents through secondary-market acquisitions. 

In the same period, Malikie Innovations, the licensing subsidiary that acquired BlackBerry’s 32,000-strong patent portfolio for $200 million in 2023, launched enforcement campaigns against Xiaomi, Acer, and ASUSTek. Acer and ASUSTek settled. Xiaomi did not.

The question the above data raises is why, despite five years of sustained volume, most organizations in the assertion path have not recalibrated their exposure assessments to match it.

The answer sits in the patent acquisition market.

Sixty Percent of Base Station Patent Acquisition Volume Went to NPEs

Between 2021 and 2025, over 1,900 patent families specifically related to base station technology changed hands. Non-practicing entities and licensing companies accounted for 58.5% of all purchases by industry category, approximately 1,162 families. 

The three most active buyers over the entire period were Malikie Innovations (415 families), Ani Acquisition Sub LLC (264 families), and Peninsula Technologies LLC (108 families).

This connects the courtroom pattern to the ownership pattern. The patents driving base station exposure are not merely being litigated; they are also being concentrated in the hands of entities built around monetization.

SEP Status of the Acquired Patents

The next question is not only who is buying these assets, but what kind of patents they are buying.

A significant and growing share of the base station patents that have transferred to non-practicing entities carry no declared SEP status. They are implementation patents, covering how base station technology is built and operated, not what was contributed to a cellular standard. FRAND commitments do not apply to them. There is no established royalty benchmark to anchor a negotiation. There is no cross-licensing offset available to a defendant whose IP position is built entirely around declared SEPs. The licensing dispute that emerges from an implementation patent assertion starts from a blank sheet, and the blank sheet almost always favors the party holding the patent.

Also Read:

The 5G Licensor with Fewer Patents Is Winning the Negotiations with this one Strategy 

Read Post

Recent transactions illustrate this clearly. The Huawei-to-Honor transfer in 2021 involved 183 patents, 172 of which were declared SEPs, a ratio of 94%. The strategic logic is transparent. Declared SEPs carry FRAND obligations but also come with an established licensing infrastructure and commercially recognized benchmarks. Honor acquired standard-essential leverage suited to cross-licensing negotiations with peers operating in the same FRAND-governed framework.

That transaction shows how SEP-heavy transfers fit within a known licensing framework. The contrast becomes sharper when the acquired portfolio falls outside that framework.

By contrast, the Corning-to-Ani Acquisition deal in 2025 involved hundreds of patents with no declared SEP status.

Every patent in that portfolio is an implementation patent. There is no FRAND framework governing their assertion, no established royalty rate to anchor negotiations, and no cross-licensing offset available to a defendant whose IP position is built solely on standard-essential patents. The Ofinno-to-Peninsula Technologies transfer in the same year moved 108 families, each with a declared SEP count of zero.

The consequence of that shift is already visible in recent filings.

When we see the consequences, we found that “Peninsula Technologies LLC” has filed cases, i.e., 2-25-cv-00386 (defendants – Dish Wireless, LLC d/b/a Boost Mobile, VOXX International Corp), 2-25-cv-00387 (defendants – Dish Wireless, LLC d/b/a Boost Mobile), 2-25-cv-01028 (Defendant – Chipotle Mexican Grill, Inc.) & 2-25-cv-01024 (Fortinet, Inc.).

The Policy Argument and the Asset Movement Are Pointing in the Same Direction

This base station patent acquisition is not happening in isolation. It is unfolding alongside a broader argument about where telecom licensing value should sit.

Alongside these transactions, a broader shift in thinking began surfacing at industry forums. 

At IPBC Asia in 2025, licensing strategists from Xiaomi and Oppo surfaced a paper authored by Dr. John Gong, a Chinese economics professor. It argues that base station companies should bear significantly higher SEP licensing costs, effectively proposing a rebalancing of royalty flows away from handset manufacturers and toward the infrastructure layer. That debate has not been resolved.

“What is observable, however, is the timing. The companies most closely associated with advancing that argument have simultaneously been active in transferring base station-related patents to third-party entities. Policy advocacy for higher infrastructure-layer royalties and the parallel movement of base station assets into assertion vehicles are not, on the available evidence, unconnected developments. One provides the commercial rationale; the other builds the enforcement infrastructure to capture it.”

The combined effect is a repositioning of base station patents away from the standard-based licensing frameworks that have governed SEP disputes for two decades, toward assertion strategies in which fewer established rules apply, and negotiating outcomes are considerably less predictable. For the companies on the potential receiving end of those strategies, that shift, from FRAND-constrained SEP licensing to unconstrained implementation patent assertion, represents a categorically different risk profile than the one their current exposure assessments are built to address.

Once the market and policy context are clear, the next step is to look at the specific patents repeatedly used in litigation.

5 Most Frequently Used Base Station Patents In Litigation 

Understanding which patents drive the highest litigation volume provides a practical lens on where assertion risk concentrates.

Cases by status of the top 5 most litigated patents

The five most-asserted base station patents span fundamental physical-layer technologies, all of which are currently held by Fleet Connect Solutions LLC. US7260153B2 (Fleet Connect Solutions LLC) pertains to SVD-based MIMO processing, including the calculation of singular values and pre-equalizer matrices, a core physical-layer algorithm used in MIMO base stations.

US7058040B2 (Fleet Connect Solutions LLC) and US7656845B2 (Fleet Connect Solutions LLC) focus on data transmission over overlapping media; the latter explicitly claims base station-side allocation functions relevant to radio resource management and scheduling. 

US6633616B2(Fleet Connect Solutions LLC) describes a pilot phase error metric hardware block for OFDM receivers, directly relevant to base station hardware performance. US6549583B2 (Fleet Connect Solutions LLC) discloses a method of pilot phase error estimation, a fundamental PHY-layer signal processing function required for standard-compliant operation in LTE, 5G, and Wi-Fi systems.

Also Read:

Why 500 Relevant LTE-M Patents Beat 2,000 Declared LTE Families Every Time

Read Post

From a portfolio perspective, in MIMO channel crosstalk parameter determination, companies such as Nxgen LLC, Sony, and Qualcomm hold a significant number of patents. In base-station scheduling, Huawei, Nokia, and Ericsson appear to have strong portfolios. In phase tracking using pilots, Qualcomm, Huawei, NTT, and Apple hold a substantial number of relevant patents.

Their current litigation status shows that these patents are not just historically important; they remain active enforcement assets.

Current litigation status of these patents 

At present, these patents are involved in more than five active litigation cases.

The ongoing litigation involves companies that manufacture telematics devices, GPS tracking units, and IoT hardware that rely on cellular or wireless transmission, including CalAmp Corp., PowerFleet Inc., UAB Xirgo Global, Xirgo Holdings Inc., Xirgo Technologies LLC, Eroad Ltd., Coretex Ltd., and Masternaut Limited. The litigation also extends to vehicle manufacturers with built-in connected telematics and wireless transmission units, such as PACCAR Inc. and Rivian Automotive LLC.

In addition, companies manufacturing consumer technology devices, including cameras, dashcams, and audio equipment with integrated wireless transmitting units such as Wi-Fi, Bluetooth, or cellular connectivity, have also been litigated using these five patents. Examples include Bose Corp., Nikon Corp., OM Digital Solutions Corp., and Olympus Corp.

Taken together, the litigation data, reassignment activity, SEP-status split, and active case record all point to the same conclusion.

The Patents That Will Drive the Next Five Years of Base Station Litigation Are Already in NPE Hands

The companies that will face the next wave of base station assertions may never have considered themselves licensing targets. They do not manufacture handsets. They do not hold weak SEP positions. Some hold no SEP exposure at all. What they share is that their products implement base station technology, and the patents now consolidated in NPE hands cover implementation, not standardization.

A robust declared SEP portfolio offers no defense against a claim that sits entirely outside the standardization framework it was built to address.

The practical implication is not about litigation readiness. It is about timing. By the time a complaint is filed, the strategic options available to a defendant have already narrowed considerably. The transfer records that would have signaled the risk were publicly available months, sometimes years, earlier, visible to those systematically watching the secondary market, invisible to those who were not.

At this point, the questions that matter are specific. 

• Which NPE currently holds patents that read on a specific product line? 
• Which technology area within a company’s stack carries the highest concentration of NPE-owned implementation patents? 
• Which of those patents have an AI-generated risk assessment already mapped to a specific product?
• If a new NPE has entered the base station space in the last quarter, which companies are in its immediate assertion path, and which products are at risk?

The GreyB NPE Risk Dashboard is built to answer those questions before they become urgent.

It tracks daily the patents acquired by NPEs post-2022, maps them to companies and specific products at potential risk, identifies the technology areas with the highest concentration of NPE ownership, and surfaces AI-generated risk assessments at the individual patent level. That is why the risk assessment has to begin before a complaint is filed.

Request a demo of the NPE Dashboard by filling out the form below.

Methodology: 

The conclusions in this analysis are drawn from a structured, multi-stage examination of US patent litigation and secondary market transfer data spanning 2021 to 2025. The methodology is worth stating explicitly, because the credibility of the counterintuitive findings it produces depends entirely on the rigor with which the underlying dataset was constructed.

The starting point was a corpus of 3,224 litigation cases filed in the United States over the five-year period, encompassing approximately 16,919 patents involved across those proceedings. That dataset was not treated as a single body of telecom litigation. It was filtered in two distinct stages designed to isolate base station-specific exposure with a precision that broad litigation surveys do not typically achieve.

The first stage tagged every patent involved in those cases to determine whether it carried telecom relevance, separating the telecom-related docket from the broader patent litigation environment. That filter identified 19.9% of cases as telecom-related, establishing the universe from which base station exposure was then measured.

The second stage applied a more exacting standard. Each patent within the telecom-relevant cases was examined at the claim level, specifically, whether any independent claim disclosed a base station claim or described a method performed by a base station. Patents whose claims focused exclusively on the user equipment side were excluded. This is the methodological decision that most distinguishes this analysis from surface-level litigation reviews, which frequently conflate user equipment and base station claims within the same telecom category and consequently understate or mischaracterize where assertion risk actually concentrates.

The two-stage tagging process found that, on average across the five-year period, 52% of all telecom-related cases involved base station claims, a figure that holds consistently across each year of the dataset rather than being skewed by a single anomalous period.

Because NPEs were identified as the dominant plaintiff class within the base station litigation docket, the analysis was extended to incorporate patent reassignment data for the same five-year window. That reassignment dataset, covering over 1,900 base station patent families, provides the secondary market context that explains why litigation volume has remained structurally elevated and where the assertion pipeline for future enforcement campaigns currently stands.

The post Why Telecom Companies Need to Watch Base Station Patent Transfers More Closely in 2026 appeared first on GreyB.

Sakshi’s team hit a bit of a wall during an ingredient-scouting project for a leading chocolate company. The key question was simple but tricky:

“How do you even test ingredients that aren’t finished product yet?”

The team needed a way to make ingredient performance more practical, visible, and easier for the client to evaluate.

Making Ingredient Performance Easier for the Client to Compare

Approach 1: Testing Real Products That Use Similar Ingredients

Instead of testing ingredients in isolation, the team identified and analyzed existing chocolates and confectionery products that already used similar ingredients. A list of products was prepared for procurement. Most were sourced from India, while a few came from the US. Shikhar and Rohit brought back the US products during their return trip. Another Greybian, Rishabh, who was in Europe, was also brought into the process to scout relevant products there, taste them, and share his reviews.

This helped them:

• Predict how the ingredient might translate into actual taste
• Understand flavor profiles, texture, and overall appeal
• Identify what works and what does not
• Benchmark against existing successful and unsuccessful products
• Give the client a much more tangible sense of how the ingredient could perform in the real world

In short, the team turned something abstract into something they could actually taste, compare, and respond to.

Approach 2: Building a Consumer Survey Without Relying Only on Agencies

The usual route would have been to pay third-party agencies. However, the team decided to take a more creative route.

Here’s what they did:

• Referenced existing third-party surveys and reports already available online
• Created a survey tailored to this project
• Shared it with friends and family in the US
• Posted it on Reddit to get broader, organic responses

Now, one concern was obvious: Would Reddit responses affect the quality of the data?

The team accounted for this by adding a simple filter:

• First question: “Are you a US respondent?”
• If “No,” the form ended immediately
• If “Yes,” the respondent proceeded

It was not a perfect system, but it was an effective way to keep the responses relevant.

Over the course of the week, the team procured and tested several ingredients in their commercially available forms. Each product was evaluated for taste, mouthfeel, odor, aftertaste, solubility, and repurchase intent. These findings were then cross-referenced with what consumers had reported online.

All this was done even though neither product testing nor consumer surveys were part of the original proposal.

Product Testing Became a Strategic Priority, Not a One-Time Add-On For The Client

The team presented this on a client call yesterday. Before the deck was even opened, one of the client team members mentioned that she had reviewed the update beforehand.

Her first words on the call were:

“I saw you guys did product testing — it is quite impressive. I’m curious to know, are you going to do this for all the ingredients?”

When a client asks for more of something before the team has even walked them through it, that is more than positive feedback. It is a signal.

It showed that the client was no longer seeing product testing as a one-time add-on. They were beginning to see it as a lens that could be applied consistently across the project.

Matcha Emerged as a Clear Segment Signal in Product Testing

For most ingredients, the testing confirmed what secondary research had already suggested. But for a few, it added a layer of insight that online data could not have surfaced.

1. A Color Concern Turned Out to Be Brand-Specific

For one ingredient, online reviews flagged color as a recurring complaint. However, the team did not experience this issue during testing.

This showed that the color concern was brand-specific, not inherent to the ingredient. For client, that meant it was a manageable formulation variable, not a red flag against the ingredient itself.

Without physical testing, this distinction would have been missed.

2. Acceptable Online Reviews Did Not Match the Actual Taste Experience

For another ingredient, consumer reviews online were broadly acceptable. But when the team tested the product, the flavor felt overpowering, artificial, and extra sweet.

The team also found internal color inconsistency. Some pieces were pink, while others were yellow, despite having the same flavor. No online reviewer had flagged this.

The team’s consensus was clear: they would not buy it again.

This insight would have been missed entirely if the team had relied only on secondary data.

3. Matcha’s Bitter Edge Limits Mass Appeal, but Appeals to Dark Chocolate Lovers

Matcha was the most interesting case.

Online reviews for matcha were mixed, and the team’s response was also mixed. At first, this looked like a simple confirmation of what was already available online.

But when the team looked more closely at who liked it and who did not, a pattern emerged.

The people who enjoyed the pleasant bitterness of green tea or matcha were also those who liked dark chocolate or preferred high-caffeine beverages. They already had a palate for pleasant bitterness.

This is not something an online survey would have revealed because reviewers usually do not describe themselves that way.

So, what looked like a generic “mixed review” online became a segment signal. Matcha’s bitterness may not be a problem to fix. It may be a targeting lever.

This was especially relevant because the client already played in the dark chocolate space.

Discover Ingredients That Match Your Product’s Vision

Most companies spend 18-24 months developing a single new product, with ingredient sourcing consuming nearly 40% of that timeline.

GreyB’s Ingredient Scouting service helps food, beverage, nutraceutical, and consumer goods teams identify promising ingredients, evaluate their technical and market relevance, benchmark available alternatives, and shortlist options that are more likely to work in real product pipelines.

Need a sharper shortlist for your next formulation or innovation pipeline? Schedule a consultation with our experts today.

The post When Ingredient Testing Was a No-Go, GreyB Tested the Products Instead appeared first on GreyB.

Despite years of global investment, the water crisis is far from solved. The UNSD report shows that 2.2 billion people still lack access to safely managed drinking water, while 3.4 billion live without safely managed sanitation. The same report also highlights that only 56% of domestic wastewater is safely treated. At the current pace, the world will not achieve sustainable water management until at least 2049.

In India, urban centers generate 72,368 MLD of sewage, yet installed treatment capacity is only 31,841 MLD (~50%). And only 28% of that sewage actually gets treated. The remaining 72% flows untreated into rivers, lakes, and aquifers. These figures show the scale of the water waste management crisis.

Startups like Avior Aqua are trying to address this problem by improving one of the most energy-intensive and inefficient parts of water treatment: aeration. 

To better understand how they are doing it, we spoke to Sushant Das, Founder & CEO of Avior Aqua. This article contains notable highlights from our entire conversation.

This interview is part of our exclusive Scouted By GreyB series. Here, we speak with the founders of innovative startups to understand how their solutions address critical industry challenges and help ensure compliance with industry and government regulations. (Know more about startups scouted by GreyB!)

Also Read:

The solution to toxic algae was hiding in a steel mill and this startup found it!

Read Post

This startup fetches 99% of valuable lithium from polluted wastewater!

Read Post

“The next big reason for the next war would be water.”

– Sushant Das

Sushant Das is the Founder & CEO of Avior Aqua, which he co-founded with engineering, healthcare, IT, and marketing professionals, united by a shared concern for India’s water crisis. Overseeing product development, engineering, and operations, Sushant has steered the company’s mission to make non-potable water reusable through nanobubble technology.

Under his leadership, Avior Aqua has successfully piloted lake rejuvenation in Mira Bhayandar, conducted pilots at Wankhede Stadium and Pirana STP in Ahmedabad, won the MSW 2025 award, and earned selection in India’s national Amrut 2.0 “Pitch to Pilot” challenge.

Avior Aqua is making water treatment more efficient with nanobubbles

Avior Aqua generates nano-sized bubbles of oxygen, air, or ozone compared to the larger bubbles used in the water treatment industry. These small bubbles remain dissolved longer in water, thereby improving gas transfer. 

This technology can be applied across sewage treatment plants, industrial effluent plants, lake rejuvenation, swimming pools, agriculture, stadium turf, and golf courses.

The vision of Avior Aqua with this technology is to improve oxygen transfer in water, reduce chemical use, and make existing water treatment systems more efficient without always requiring a complete infrastructure rebuild.

What does Avior Aqua’s technology do in simple terms, and how is it different from traditional aerators?

Sushant: In simple terms, we make bubbles extremely small. Traditional aerators create larger bubbles that rise to the surface and escape within seconds. As a result, only about 5–10% of the gas typically dissolves in water, and the rest is wasted. This is why aeration machines often need to run 24/7.

Our technology replaces conventional aeration with nanobubbles. Each bubble is smaller than 150 nanometers, so it can remain in water for months when kept in a controlled environment. Because the bubbles are so small, we achieve much higher gas dissolution, which makes the treatment process more efficient.

The core technology is the same across applications, but the equipment varies by use case. A sewage plant, swimming pool, lake, and industrial effluent plant all need different setups, even though the underlying nanobubble principle remains the same.

How can this technology improve sewage treatment plants?

Sushant: Sewage treatment depends heavily on oxygen because biological oxygen demand, or BOD, is one of the key parameters. If a sewage treatment plant uses normal air for aeration, we can replace it with oxygen nanobubbles. By changing that part of the system, the same plant can often treat more water while maintaining the same power consumption.

For example, if a plant is designed for 100 million liters per day, replacing conventional aeration with nanobubbles can help increase the plant’s capacity significantly. In some cases, we can double capacity without doubling the infrastructure. The treated water quality also improves, making it more suitable for reuse.

This matters because most treatment plants rely heavily on aeration, and much of the gas being pumped into the water is wasted. We are trying to make that gas transfer more efficient, so the same treatment plant can do more with less.

How do you maintain consistent performance when water quality varies so much?

Sushant: We first look at the existing water parameters and what the client expects after treatment. Based on that, we decide on the equipment, dosage, and supporting systems. Nanobubbles are usually part of a larger treatment approach, so sometimes we also need filters, screens, probiotics, or other solutions.

For sewage water, we may use a general thumb rule for biological demand, but if the incoming sewage is worse than expected, we increase the dosage. If the water is better than expected, we can either reduce the dosage or use the same equipment to increase the treatment capacity.

So the technology is not a one-size-fits-all machine. The core remains nanobubbles, but the treatment design is customized to the site.

What would an industrial client see in terms of cost and ROI?

Sushant: It depends on the application. Some industrial clients use our system for wastewater treatment, but others use it to optimize their actual production process. For example, in textiles, water is a major part of dyeing. With this technology, we can reduce water use in that process by a large percentage.

If the system is only for wastewater treatment, the capital cost may be around five to ten lakhs, depending on the size and complexity. For process optimization, the range can start from around one and a half to two lakhs and go up to a few crores.

Clients do not only measure chemical reduction. They also look at power savings, operational savings, process efficiency, and the quality of water coming out. Chemical reduction is one visible benefit, but the actual ROI usually comes from a combination of many improvements.

How quickly can a new installation start showing results?

Sushant: Once the equipment is installed, results can usually be seen within a week or two. The manufacturing and delivery of the equipment can take around one to two months, but the treatment impact after installation is quite fast in most use cases.

Lakes are different. A lake is like a large living organism, and every lake behaves differently. Lake rejuvenation is more dependent on nature, so we generally suggest a minimum of three months before expecting visible changes.

For sewage plants, swimming pools, farms, or industrial systems, the results are much faster. In many cases, the difference is almost instant once the system starts operating.

Where is Avior Aqua seeing strong use cases beyond wastewater?

Sushant: Agriculture is one important use case because we increase oxygen levels in water. When plants receive oxygen-rich water, we have seen better output, stronger plants, and better resilience. In one pilot near Navi Mumbai, basil grown during summer survived high temperatures while the control batch failed.

We are also working with stadiums and golf courses because root oxygenation is a major challenge in turf management. Stadiums often have brown patches, and sometimes fake grass is used to cover them during matches. By improving oxygen availability at the root level, we can help maintain healthier turf.

We are also developing smaller products for household use. Today, our machines are still too large for individual homes, but we are working on a compact version that could eventually fit into a domestic setting.

What are your expansion plans for the next five years?

Sushant: For the next five years, our focus is on the environment. That means sewage, effluent, lakes, water bodies, and wastewater. We want to go deep into these areas and maximize our impact there before spreading ourselves too thin.

We are also researching healthcare and wellness applications. I think within the next two years, we may enter healthcare in some form, but the main focus will remain on environmental applications.

Nanobubbles have many potential use cases, but one company cannot cover everything at once. We want to focus on the areas where we can create the most immediate impact.

How do you protect the novelty of your technology?

Sushant: We have filed patents, and we are constantly filing more. Protecting our work is important, but I do not want nanobubble technology to remain limited to only Avior Aqua. Even in the next 50 years, one company will not be able to cover all possible applications of nanobubbles.

The nanobubble community is very collaborative. A lot of companies and researchers openly publish their work, and that helps everyone. International research done by others also helps me in business development because it builds trust in the category.

So yes, we are protecting what we build, but we are also open to collaboration. If someone wants to use our technology for something like drug delivery, and it makes sense, we would be open to working with them. The more this technology spreads right now, the faster it will be adopted over the next five years.

Meet our Interviewer – Anusha Srivastava, Senior Research Analyst

Want to find other scalable startups working on water treatment technologies? Please fill out the form below to contact our experts.

The post This Startup Replaces Chlorine with Oxygen-rich Nanobubbles for Cleaner Water appeared first on GreyB.

A licensor walks into a Cat-M1 negotiation citing 2,000 declared LTE families. The implementer across the table does not dispute the number. Instead, they ask a different question: how many of those families cover features does a smart meter actually use?

That question is the new negotiation. And most licensors are not ready to answer it.

The Declared LTE Pool Is Three Times Larger Than the LTE-M Implementation Envelope

The ETSI IPR database records 29,416 declared LTE patent families as of March 2026. Approximately 80 percent, 23,696 families, carry at least one alive-granted member across at least one jurisdiction. The pool has commercial weight. It is enforceable. And for IoT licensing, it is largely the wrong unit of analysis.

LTE-M devices, smart meters, asset trackers, industrial sensors, and connected health wearables do not implement the full LTE feature set. They operate within a constrained technical envelope: maximum 1 Mbps downlink, a single receive antenna, 1.4 MHz narrowband operation, half-duplex FDD, and low-power procedures including PSM and eDRX. The gap between what LTE declares and what LTE-M devices deploy is not minor. Feature-level analysis of the 29,416-family pool shows that only 56 percent of declared LTE families appear relevant to LTE-M implementations at all. The remaining 44 percent maps to features outside the Cat-M1 capability envelope entirely.

Within that 56 percent, the separation matters further. Foundational families, those covering features mandatory for LTE-M compliant operation, account for 32 percent of the total declared pool. Optional families, relevant to some deployments but not universally implemented, account for 24 percent. 

A licensor holding 2,000 declared LTE families, therefore, enters a Cat-M1 negotiation with roughly 640 foundational assets before any quality or jurisdictional filtering begins. The implementer’s question is not unfair. It is arithmetically correct.

ETSI Declarations Were Never Designed to Sort by Device Category

The structural problem is not a bad-faith declaration. It is that ETSI’s IPR database was built to record essentiality claims against standards, not against device categories. A patent declared against TS 36.331 may cover RRC signaling that every LTE-M device must implement. Another patent declared against the same specification may cover carrier aggregation procedures that Cat-M1 devices never execute. Both entries appear identically in the database. Neither carries a flag distinguishing foundational LTE-M relevance from full-LTE-only applicability.

This creates the asymmetry that implementers are now exploiting. When a licensee asks whether a portfolio is relevant to their specific device, the declaration database cannot answer. The licensor who relies on that database and leads with total family count has already ceded the evidentiary high ground. As one practitioner observation captured in recent licensing commentary notes, when licensors cannot demonstrate transparency in mapping declared assets to implemented features, implementers have standing to contest FRAND validity on proportionality grounds.

The three questions that implementers are now posing in Cat-M1 discussions encode this shift directly: why should full LTE rates apply to a device implementing only a subset of LTE; which licensors actually contribute to LTE-M functionality rather than LTE generally; and are all portfolios equally relevant once the device lens is applied? The third question is the most consequential. Portfolios that look comparable at the declaration level diverge sharply once carrier aggregation, MIMO, 256-QAM, and URLLC features are removed from scope.

“Portfolio size gets attention and may open the conversation, but when an implementer asks for proof, relevance closes it. A licensor with 500 foundational LTE-M families carries more defensible leverage than one with 2,000 broadly declared assets that don’t survive device-level scrutiny.” — Aman Kumar, IP Consulting, GreyB.

Foundational Coverage Is Concentrated in Release 8, Not Release 13

The counterintuitive finding in the data concerns release attribution. Release 13 is where eMTC was standardized, and it contributes 788 foundational families to the LTE-M relevant pool. That is a commercially meaningful number. It is also dwarfed by what Release 8 contributes: 6,965 foundational families.

The reason is architectural. LTE-M does not replace the LTE baseline. It operates on top of it. The procedures an LTE-M device relies on for network discovery, attachment, authentication, paging, and basic data transfer were defined in the earliest LTE release. Release 13 commercialized eMTC and introduced Cat-M1-specific enhancements. Release 8 built the foundation that those enhancements depend on.

The licensing implication is two-directional. Licensors who focus exclusively on Release 13 assets as their LTE-M story are leaving the majority of their foundational coverage undefended. Their early LTE families may carry stronger Cat-M1 relevance than their later IoT-specific assets, simply because the core network and physical layer functions defined in Release 8 appear in every LTE-M device implementation.

Conversely, licensors who assume that all LTE Release 8 families are automatically relevant to LTE-M are overstating. A Release 8 patent covering carrier aggregation preparation procedures is still outside the Cat-M1 envelope regardless of its age.

The analytical unit is not the release date. It is the feature. The release date is a starting point for prioritization, not a substitute for mapping.

“The strongest foundational relevance in LTE-M sits in the basic LTE functions that allow devices to discover the network, attach, conserve power, and transfer data. A smart meter does not need the same throughput profile as a smartphone, but it absolutely needs reliable attachment, coverage, paging, and signaling. Those procedures trace back to Release 8, and they are where claim-level mapping produces the most defensible licensing position.” — Nripdeep Singh, Senior IP Analyst, GreyB

Ownership Concentration Narrows the Benchmark Set, but Relevance Still Decides Leverage

The foundational and optional LTE-M relevant pool is not evenly distributed across the industry. The top 10 assignees collectively hold approximately 62 percent of those families. Traditional telecom incumbents dominate, and Chinese entities hold a substantial share within the leading group. For licensing negotiations, this concentration defines the competitive context: licensors will be benchmarked against a small number of players, not against a diffuse field.

Jurisdictional spread does not alter this dynamic. Patent families in the relevant pool carry coverage across all three major enforcement jurisdictions. The US holds 23,209 families with foundational representation; EP covers 19,056; CN covers 21,581. The geographic profile is broad and roughly balanced. A strong foundational family is likely to carry actionable coverage in each major market.

But the jurisdictional data underscores rather than resolves the core challenge. A granted family in the United States, Europe, and China is a strong asset only if it maps to a feature the licensed device implements. Jurisdictional breadth is a multiplier applied to relevance. Applied to an excluded feature, it multiplies nothing.

The competitive question for any licensor entering a Cat-M1 negotiation is therefore not whether their portfolio is globally filed. It is what percentage of their foundational and optional families covers the features that the top-10 assignees are also asserting, and where their coverage is differentiated rather than overlapping.

The Methodology That Converts Portfolio Size into a Negotiation Position

Filtering 29,416 declared families to a defensible LTE-M subset requires a structured process, not a keyword search.

The taxonomy that underlies this analysis is built from 3GPP specifications, TS 36.331, 36.213, 36.211, 36.300, 36.212, and related documents, structured into a four-level hierarchy: domain, feature, sub-concept, and technical mechanism. Each node in that hierarchy is classified against Cat-M1 capability constraints: foundational if mandatory for compliant operation, optional if supported by a subset of deployments, excluded if outside the Cat-M1 envelope entirely. 

The excluded category captures not only obvious exclusions like MIMO and carrier aggregation but also the important distinction between NB-IoT features and LTE-M features. Broad MTC references do not automatically imply LTE-M relevance. The taxonomy enforces specificity.

Patent tagging applies an AI-assisted engine that maps abstracts, independent claims, and priority context against taxonomy nodes, processing families in batches with SME validation at each iteration. The output is not a keyword-matched approximation. It is a relevance scorecard that separates foundational families from optional and excluded families at the claim level, benchmarks that scorecard against the declared pool, and produces the evidence package a licensing discussion requires.

For licensors, the output defines the subset of families that can survive implementer scrutiny and the feature coverage narrative that justifies the rate. For implementers, it provides the basis to test whether an asserted portfolio actually maps to their device’s implementation profile. For counsel, it directs essentiality review, claim charting, and validity analysis toward the assets that will carry the most weight in negotiations or disputes.

Five Questions That Determine Your Position Before the Next Meeting

Before entering any Cat-M1 licensing discussion, as licensor, implementer, or counsel, the following questions should have specific, data-backed answers:

1. Of your declared LTE families, what percentage maps to LTE-M foundational features as classified against Cat-M1 capability constraints?
2. Which technical domains carry your highest foundational density, core network and EPC, physical channels and reference signals, or RRC and radio resource management, and what does that profile imply for the device categories you are targeting?
3. How does your foundational share compare to the top-10 assignees that collectively hold 62 percent of the relevant pool?
4. Of your Release 8 families, how many map to LTE-M baseline procedures versus features that fall outside the Cat-M1 envelope?
5. When a counterparty asks which families cover PSM, eDRX, MPDCCH, or narrowband reference signals, can you produce claim-level mapping in the meeting rather than promising to follow up?

If any of these questions does not have a ready answer, the negotiation begins with an information asymmetry that the implementer will use.

An LTE-M relevance scorecard, by patent family, identifies which assets in your declared LTE portfolio are foundational to Cat-M1 devices, optional, or outside the implementation envelope. This provides you with a defensible rate narrative and gives implementers a basis to evaluate asserted portfolios against their actual product.

Fill the form to receive your portfolio-level LTE-M relevance scorecard

The post Why 500 Relevant LTE-M Patents Beat 2,000 Declared LTE Families Every Time appeared first on GreyB.

According to the Global Burden of Disease Study, cases of Urinary tract infections increased by more than 66% between 1990 and 2021. The problem is not just scale. Standard urine culture, still considered the gold standard for UTI diagnosis, often requires 24–48 hours for results. Antibiotic susceptibility information can take even longer. That delay forces clinicians to start UTI treatment before knowing the exact pathogen or resistance profile. This increases the risk of ineffective treatment and unnecessary antibiotic use.

GeneSys Bio is solving this gap with an AI-powered, portable genetic testing platform that detects UTI-causing pathogens and antibiotic resistance markers in 15–45 minutes. 

To better understand how they are doing it, we spoke to Andrea Faviere and Enrico Di Oto, CEOs of GeneSys Bio. This article contains notable highlights from our entire conversation.

Also Read:

4D Biomaterials Making New Generation of Resorbable Medical Devices – Insights from Phil Smith, CEO

Read Post

How Personalized Medical Device Startups Are Shaping Healthcare in 2025

Read Post

This interview is part of our exclusive Scouted By GreyB series. Here, we speak with the founders of innovative startups to understand how their solutions address critical industry challenges and help ensure compliance with industry and government regulations. (Know more about startups scouted by GreyB!)

“We wanted to have a solution that was from sample to results, not just chemicals, not just advice, but we want to be able to just own all the process.”

– Andrea Faviere

Andrea Faviere is the Co-Founder and CEO of GeneSys Bio. With an MBA/EMBA background in company management and a decade of experience in business development, marketing, and sales, Andrea leads the company’s commercialization strategy and market positioning.

Under his leadership, GeneSys Bio has advanced its rapid diagnostic platform from concept toward validation and regulatory readiness, while attracting support from institutions such as Cassa Depositi e Prestiti, OaCP IE Ltd, and AC75 Startup Accelerator. 

His role is to bridge the company’s molecular diagnostics innovation with real clinical adoption, making advanced testing faster, easier to use, and more accessible at the point of care.

Dr. Enrico Di Oto is the Co-Founder and COO / Investor Relations lead at GeneSys Bio. With a background spanning biology, clinical pathology, oncology, and experimental pathology, Enrico brings 18 years of experience in hospital diagnostics to the company.

Before GeneSys Bio, he co-founded OaCP, Oncology and Cytogenetic Products, where he helped commercialize DNA diagnostic technologies across multiple countries. 

At GeneSys Bio, Enrico plays a key role in shaping the company’s scientific and operational strategy, helping turn complex molecular biology into a faster, portable, and clinically reliable testing platform.

GeneSys Bio Tackles Antibiotic Resistance at the Point of Care

GeneSys Bio is developing an AI-powered portable genetic testing platform for detecting infectious diseases and antibiotic resistance. The company is starting with urinary tract infections because they are widespread, frequently tested, and closely linked to antibiotic misuse. Its first solution combines a disposable kit and a compact device, roughly the size of a toaster, to detect the five most relevant UTI pathogens and two antibiotic resistance families.

The company aims to give clinics, pharmacies, and hospitals a faster way to choose the right treatment without waiting days for central lab results. This matters even more as the WHO’s 2025 antibiotic resistance surveillance report analyzed more than 23 million confirmed cases of bloodstream infections, UTIs, gastrointestinal infections, and urogenital gonorrhea, highlighting the growing need for improved testing capacity.

What is GeneSys Bio doing, and what problem are you trying to solve?

GeneSys Bio is developing an AI-powered portable genetic testing platform for detecting pathogens and antibiotic resistance. We started with urinary tract infections because they are one of the most common infections worldwide, and the current diagnostic workflow is still too slow for many real-world clinical decisions.

With our device and kit together, we can detect the five most relevant UTI pathogens and two families of antibiotic resistance in 15–45 minutes. This means doctors, clinics, pharmacies, and other care settings can identify the infection and understand resistance risk much faster than with traditional testing.

Our goal is to help clinicians define a proper treatment without the fear of false results or antibiotic resistance. We are not only trying to diagnose the infection; we are trying to support better antibiotic decisions from the start.

How does your technology reduce a test that usually takes much longer to just 15–45 minutes?

The simple explanation is that we empower the DNA testing reaction through our proprietary chemistry and device design. Our kits carry the biomarkers needed for detection, and our device is built to run the reaction in a very compressed timeframe.

The device also includes an in-house algorithm that helps interpret the result and check reliability. So the speed comes from the combination of chemistry, hardware, and software, not from one single feature.

I usually call it the magic of science, but behind that magic, there is a lot of design work. We are taking a molecular biology approach and making it faster, smaller, and easier to use.

Why did you start with UTIs, and where else could the platform be used?

We started with urinary tract infections because the need is very clear. There are hundreds of millions of cases every year, and many countries need faster and more reliable testing to reduce unnecessary antibiotic use.

But our platform is not limited to UTIs. We designed it with a broader One Health vision, meaning it can eventually move across human health, animal health, and environmental testing. In human health, we could address sexually transmitted infections, prosthetic joint infections, malaria, tuberculosis, and other infectious diseases.

The same approach could be applied to infectious diseases in veterinary settings. In environmental testing, we see strong potential for waterborne pathogens, which are a major issue not only in developing countries but also in developed markets.

What makes GeneSys Bio different from other rapid diagnostic solutions?

We built everything around the user experience. Our aim is that almost anyone can run the test. You take the sample, drop it into a vial, place the vial into the device, press start, and read the result.

The AI and algorithmic part is important because it helps check the process and make sure the result is reliable. We do not want users to worry about interpreting uncertain results or managing complex lab steps.

Another important difference is that we combine pathogen detection with antibiotic resistance detection. Many solutions focus only on identifying the organism, but for UTIs and many other infections, knowing resistance is critical for choosing the right treatment.

Where does your strongest innovation sit: software, hardware, or chemistry?

We worked on every part of the solution. We came into this space with previous experience in cancer DNA diagnostics, and during COVID, our team developed a PCR kit that could detect COVID and three major variants in one 45-minute test.

From there, we looked at the market and understood that innovation was needed across the whole workflow. We shrank PCR into a portable device, developed proprietary chemistry for the kit, and built an algorithm behind the device.

We expect to protect the hardware with patents. We have a patent around the UTI kit, and the algorithm will be protected through copyright. The real strength is that we own the process from sample to result, instead of offering only one isolated component.

How accurate is the GeneSys Bio platform?

Right now, we have tested both the device and the kit in our own lab, and we have reached 99% sensitivity. Since we are working with PCR and molecular biology standards, we expect a high level of accuracy.

The next important step is third-party validation. We are preparing validation tests so that external data can confirm the performance in an independent way.

For us, this is essential. Speed is useful only if the result is reliable. We want clinicians and users to trust that a fast result is still a safe result.

What regulatory path are you pursuing?

Since we are based in Europe, our first regulatory path is IVDR certification. We are targeting a Class C self-test pathway, which is one of the higher levels of certification.

We know this is a long road, especially for a medical diagnostic device. The validation work we are doing is part of that regulatory strategy.

At the same time, we are exploring ways to reach the market earlier with different configurations or customized versions of the solution. But for clinical use, we know certification is necessary, and we are already working toward it.

How do you balance speed with patient safety?

Our approach is strictly clinical. Even if there are ways to accelerate market entry, we want to comply with the highest safety expectations because the answer we provide can influence treatment.

Today, clinicians often face a tradeoff. They can have rapid data with lower accuracy, or accurate data with a longer waiting time. We want to combine both speed and reliability.

That means we need deep validation and strong clinical studies. The patient should receive the right answer to a critical question, and we need to be sure first for ourselves that the data is reliable.

How does GeneSys Bio fit into existing clinical workflows?

It depends on the workflow of each hospital, clinic, or facility. Some clinics outsource testing to central labs. Others have internal labs. Each setting has different logistics, timing, and responsibilities.

Because our solution is portable and easy to use, it can be adapted to many workflows. It could support emergency use, screening, internal lab testing, or other cases, depending on the customer’s needs.

Our business model also supports adoption. We plan to use a razor-and-blade model, where the device can be provided with a minimum recurring order of kits. That helps reduce upfront cost and makes adoption easier for clinics and hospitals.

What are your expansion plans beyond the first UTI kit?

UTIs are our starting point, but the platform is much broader. We want to address other infections in human health, including STIs, tuberculosis, malaria, and prosthetic joint infections.

We also want to expand into animal health and environmental testing. Waterborne pathogens are especially interesting because they are a serious problem in many parts of the world.

The long-term vision is a portable genetic testing platform that can move across different One Health applications. We are starting where the clinical need is urgent, and the market is ready, then expanding from there.

Meet our Interviewer – Shabaz Khan, Marketing Manager at GreyB

Shabaz Khan, Marketing Manager

Shabaz, is a seasoned marketing manager and leads the Scouted By GreyB. With a decade of experience, he specializes in delivering critical insights to Innovation leaders, R&D, and IP teams about evolving tech landscapes, innovation trends, and emerging breakthrough startups. Shabaz excels at aligning research data with business needs and developing strategies to solve innovation challenges. His leadership and problem-solving skills make him a valuable asset in R&D and IP research.

Want to find other scalable startups working on healthcare technologies? Please fill out the form below to contact our experts.

The post This Startup Created a Portable Test That Could Replace Multi-day UTI Lab Workflows appeared first on GreyB.