Deep Tech Brief #02

Welcome to this week's issue of Deep Tech Brief.

Every Saturday, we break down deep tech companies building genuinely hard things, in plain language. No jargon, no hype.

This week's thread is complexity that refuses to sit still.

Three European teams, three unrelated problems, one shared enemy: the real world is messier and more changeable than the tools we use on it assume.

Standard AI freezes the moment the world shifts. Conventional filters miss what is dilute and mixed with everything else. Chronic care built around the occasional clinic visit leaves patients alone the rest of the year.

Each company below is rebuilding the tool to fit the mess.

Start with the part everyone has heard of. PFAS, the "forever chemicals," are now in drinking water worldwide and are linked to cancer, hormonal disruption, and immune problems. The catch is they are hardest to remove precisely when it matters most: highly diluted, floating in a stream full of other contaminants. Conventional options like activated carbon and ion-exchange resins get expensive and inefficient in exactly those conditions, and they often just move the problem into a new toxic waste stream.

Now flip the same problem over. Those messy industrial streams also carry valuable metals out the door. Palladium trades around €38,000 to €40,000 a kilogram, and as CEO Rhea Machado points out, losing even 10 parts per million across thousands of cubic meters of water adds up fast. PFAS, she notes, is a hidden problem: nobody walks in with one.

Porelio's answer is a material called FOMS, functionalized ordered mesoporous silicas. Picture a silica sponge shot through with uniform, nanometer-scale tunnels tuned to grab one specific molecule or metal and let the rest pass. The chemistry is not new: it sat on lab benches for roughly thirty years because nobody could produce it affordably at scale. Porelio, spun out of TU Berlin, says its patented green-chemistry synthesis finally changes that.

The edge is the combination: selectivity to catch what blunt filters miss, production cheap enough to actually deploy, and a dual use that turns a compliance cost into a revenue stream by recovering the metal it removes.

Where they are: Porelio was founded in 2024 by Rhea Machado (CEO), Javier Silva Mora (CTO), and Nikol Michailidou (CPO). This week it announced an oversubscribed €2.4 million pre-seed led by Faber, on top of roughly €2.5 million in earlier public funding, per reporting. By the company's account it has run pilots in wastewater, metal recovery, and fuel cells, holds granted patents in the US, EU, and China, and pegs the combined market for both uses at around €34 billion.

Worth watching, because the technology is real and the funding is fresh. The honest risk is the one every materials startup faces: this all has to survive the jump from kilograms in a pilot plant to tons a year in the field.

Learn more: porelio.com

Today's large language models are, as kausable co-founder Gregor Ramien frames it, geniuses with amnesia. They learn the world once, during training, and then their knowledge is frozen. Ask them about something new and they are lost. That is fine for a chatbot, less so for anything at a frontier, where the world does not hold still.

This is not a fringe complaint. It is arguably the central unsolved problem in the field: Google Research published a whole new paradigm at NeurIPS 2025 aimed squarely at "catastrophic forgetting," the way a model loses old skills when it learns new ones. When the biggest labs are still writing papers on a problem, you know it is hard.

kausable is building what it calls a foundation model for dynamic, complex systems: AI meant to learn on the fly, without a full retrain every time conditions change. The pitch is a horizontal layer others build on, not a single app. The use case that lands is robotics. Today's humanoids are impressive in controlled settings and fall apart the moment you drop them into a stranger's kitchen to work an unfamiliar coffee machine. kausable wants to supply the adaptable intelligence that closes that gap. Its published research leans on causality and dynamical systems, including a benchmark called CausalDynamics developed with Pierre Gentine's group at Columbia.

The edge, if it works, is the bet itself: chasing continual learning through causal, systems-based models rather than simply making language models bigger.

Where they are: the company is based in Heidelberg, and it is early and research-heavy. By Ramien's account the team recently submitted two papers to NeurIPS, the largest AI conference. (kausable's registered managing directors are Johannes Haux and Benjamin Herdeanu.)

One to watch, carefully. Not because a product is imminent, but because the problem is real, the research is credible, and a small European team is taking a different swing at something Google has not solved either. The risk is the mirror image: this is frontier research, the field is crowded, and a foundation model for dynamic systems is still a promise, not a product.

Learn more: kausable.ai

Lymphoedema is a chronic, progressive, incurable swelling caused by a damaged lymphatic system, and it is more common than its profile suggests. Widely cited estimates put it at 140 to 250 million people worldwide, and researchers stress even that is probably an undercount, given how often it goes undiagnosed. It is also badly underserved: with few specialists, patients spend most of the year managing a lifelong condition largely alone, with little day-to-day guidance on compression, movement, and care.

Founder Mara Sorrentino knows the gap personally. Her son was diagnosed with primary lymphoedema, the rare inherited form that affects roughly 1 in 100,000 people, and the family spent three hard years learning that the right therapy is the difference between a normal life and constant trouble.

Ynone (previously LYlife) is her answer: a management hub for the conservative, day-to-day therapy that these patients live with. The app is built around symptom and swelling tracking, a therapy diary, plain-language education, and AI-supported coaching meant to spot patterns and flag risks earlier, making management a continuous thing rather than something that only happens at the clinic.

The edge is focus. By Sorrentino's account, the nearby players are indirect, such as Lipocheck, which targets lipoedema, a different though similar condition. Ynone is verticalizing on lymphoedema specifically, betting that depth in one hard condition beats a shallow tool that tries to cover several.

Where they are: this is the earliest-stage company of the week. Ynone is in Berlin's Vision Health Pioneers Incubator, is not yet incorporated, and is heading into a pre-seed, with Prajjwal Yadav as co-founder. The biggest hurdle, in Sorrentino's telling, is the one every health startup hits early: navigating EU healthcare regulation and handling sensitive patient data responsibly.

One to watch. The unmet need is real, the founder-market fit is about as strong as it gets, and the focus is smart. Temper that with the stage: an unincorporated, pre-seed team in a heavily regulated sector, with a product still to prove. The why is not in doubt; the execution is what the next year will test.

Learn more: ynone.eu

Know a deep tech company we should break down next? Hit reply and tell us. The best editions come from real tips.

This week's three founders have probably never met, but they are doing the same thing: refusing to let a static tool stand in for a moving, messy world, whether that world is a wastewater stream, a robot's kitchen, or a patient's daily care.

See you next Saturday.