Helical’s $10M Seed Wants a Virtual AI Lab for Pharma — Finally, a Wet-Lab Hall Pass

Drug discovery startups usually arrive sounding like they have personally invented both biology and urgency. Then they hand you a slide with enough gradients to qualify as a weather event and explain that a foundation model will now do for pharma what the assembly line did for Ford. I prepare myself, emotionally and metabolically, for nonsense.

And then along comes Helical, which announced a $10 million seed round on April 14, led by redalpine with participation from Gradient, BoxGroup, and Frst, plus angels including Cohere CEO Aidan Gomez and Hugging Face CEO Clément Delangue. The company calls itself a “virtual AI lab for pharma”, which is either the first sentence of a very good future or the last sentence before a compliance officer faints into a centrifuge.

Reader, I kind of like it.

The lab is virtual. The ambition is aggressively corporeal.

The pitch is not that Helical has built one magical model that will cure all disease before lunch. In fact, the company seems refreshingly allergic to the usual “our AI is the scientist now” performance. On its own site, Helical describes itself as an application layer that turns biological foundation models into reproducible discovery workflows. Translation: the models may be interesting, but the real headache is everything between a prediction and a scientist saying, “Fine, this is worth taking seriously.”

That is a much better startup thesis than “we trained a model on cells and vibes.” A lot of AI-for-science hype still behaves like the bottleneck is insight generation, when the actual bottleneck is usually messier and less cinematic: fragmented tooling, one-off notebooks, teams that cannot reproduce each other’s work, and the general institutional joy of trying to make biologists and machine-learning engineers collaborate without one side starting a sentence with “well, technically.” Helical is explicitly building for that gap.

The platform has two product surfaces: a Virtual Lab for biologists and translational scientists and a Model Factory for ML engineers and data scientists. Same data, same models, same results, fewer opportunities for two highly paid teams to generate incompatible truths in adjacent tabs. In startup terms, this is the rare AI product that sounds less like a god complex and more like a shared workspace with better abstractions.

Founders who did not all emerge from the same espresso bar

Helical says it was founded in early 2024 by three Luxembourgish childhood friends: Rick Schneider, who built tech at Amazon and later helped Celonis scale in France and Japan; Maxime Allard, who led data-science teams at IBM before pursuing a PhD in reinforcement learning and robotics; and Mathieu Klop, a cardiologist and genomics researcher.

That confidence is not entirely unearned. According to TNW’s April 14 coverage, the company is already in production with multiple top-20 global pharma companies, including a public collaboration with Pfizer on predictive blood-based safety biomarkers and another with Tanabe Pharma America on AI-driven target discovery for neurodegenerative diseases including ALS. If true, that moves this out of the “beautiful lab-demo fever dream” category and into the far more interesting category of “someone large and regulated is willing to let them near the workflow.”

The phrase “virtual AI lab” should be illegal, yet here we are

Let me be fair but honest: “virtual AI lab for pharma” is a phrase engineered in a clean room to trigger both investor curiosity and editorial eye twitches. It sounds like someone took three extremely fundable nouns, polished them to a showroom gleam, and waited for a term sheet to appear. This is the curse of startup language. Founders can have a real idea, a real product, and a real customer problem, then still insist on describing it like they are unveiling a metaverse pavilion at Davos.

And yet the more I read, the more the company’s substance keeps dragging the branding back toward credibility. Helical is selling shared infrastructure for target identification, biomarker discovery, and therapeutic design, plus the deeply unsexy virtue of making results reproducible. That matters. In science, reproducibility is the difference between progress and an expensive hallucination with nice typography.

This is why the company reminds me, in spirit, of Nomic Bio making proteomics feel unexpectedly practical. Different stack, different problem, same underlying appeal: less techno-messianism, more “here is a useful system that might help serious people get unstuck.” It also rhymes with the broader moment when medicine quietly admitted AI is already in the building, not as a TV robot surgeon, but as a slightly awkward assistant embedded in real workflows where trust matters more than spectacle.

Why investors might actually be right this time

Pharma is one of those industries where every pitch gets inflated until it briefly achieves orbit, but the market itself is not imaginary. Helical’s own announcement points to a landscape where R&D spending exceeds $300 billion annually, drug development can take more than a decade, bringing a drug to market can cost more than $2 billion, and more than 90% of candidates entering clinical trials fail. Those numbers explain why investors will keep funding anything that plausibly promises more throughput.

Also, the cap table is quietly telling a story. TNW reports that this seed round follows a €2.2 million raise in 2024 from Frst, BoxGroup, and angels including Gomez and Delangue, who returned in the new round. Repeat investors are not proof of destiny, but they are usually proof that somebody saw the first demos and came back anyway.

It helps that Helical is aiming at the application layer rather than pretending it will win the entire model war. That is a sane place to live. Ask anyone who has watched enterprise AI in the past year, or if you prefer, ask Cohere, which has also been busy explaining expensive AI abstractions to adults with budgets. Models change. Infrastructure and workflow gravity do not.

Still, a few lovingly raised eyebrows

I do not want to oversell this. Early-stage AI biotech still comes with the usual hazards: long sales cycles, difficult validation, the possibility that pharma buyers will nod enthusiastically for nine months and then buy one pilot plus a branded tote bag. Helical says teams have compressed discovery timelines from years to weeks. Maybe! I hope so. But drug discovery is a graveyard of maybe.

There is also a subtle risk in building a company around being the place where scientists and ML engineers meet. It is a real problem, but it is also the kind of problem many incumbents eventually notice and start pitching as a “unified platform” with twelve enterprise reps and a terrifying procurement PDF. The startup defense has to be speed, product coherence, and actual scientific usefulness.

Then again, the startup world often loves a grand theory more than a working system. Helical at least appears to be doing the uncool version correctly: narrow enough to matter, ambitious enough to excite, and grounded enough to survive contact with customers who can spell “biomarker” without checking. That already puts it closer to actual workflow software with consequences than to another AI mood board.

Verdict: a promising little rocket wearing a lab coat

My verdict is that Helical feels like a promising little rocket. Not because “virtual AI lab” is a good phrase. It is not. It sounds like the sort of thing a consultant would say while unveiling a glowing hexagon. But because underneath the language, there seems to be a real product thesis: scientists do not need one more dazzling model demo nearly as much as they need a system that helps them test, compare, reproduce, and defend computational work inside the brutal bureaucracy of pharmaceutical R&D.

If Helical can keep translating AI from impressive prediction machine into usable scientific workflow, investors may look less like gamblers and more like people who correctly noticed where the bottleneck lives. If it cannot, it will join the grand tradition of startups that confused “the future of science” with “a very good deck.” For now, though, I’m inclined to give the founders the benefit of the doubt. They picked a huge problem, a slightly absurd category name, and a market where being useful matters more than being cool. In startup land, that is a better seed-stage formula than charisma plus fog.