Miju LabsThe single most useful thing in the competitive map is also the least visible. BioStack Platforms is executing this exact thesis, and it has been doing so for ten months.
Founded October 2025 by Sanat Mishra (CEO) and Parth Patwa (CTO). San Francisco. Seven employees. Y Combinator Spring 2026 batch. They pivoted twice — from multi-agent biology systems, then away from in-house lab operations — before landing here (YC profile).
What they sell: longitudinal clinical data — notes, labs, imaging, medications, vitals, diagnoses, follow-ups, outcomes — converted into RL training environments where AI labs can train and evaluate medical decision-making before deployment. They build the clinical tasks (diagnosis, test ordering, medication adjustment, risk flagging) and the post-training infrastructure around them: evaluations, reward functions, benchmarks. Models are "scored against outcomes, chain-of-thought reasoning, guidelines, and clinician review" (YC launch).
Traction: 17 customers and prospects, including "top AI labs" (unnamed), and one reference to delivering data under "a six-figure contract." They are hiring a Clinical Data Lead at $120K–$200K and a Founding Research Engineer (RL/Reasoning) at $200K–$350K.
A Clinical Data Lead at $120–200K and an RL research engineer at $200–350K. The expensive hire is the environment builder, not the clinician. That is the same ordering Mercor revealed by acquiring Sepal and Deeptune for environment-construction capability rather than for expert networks — and it is the strongest available evidence about where the scarce skill actually sits. See What actually gets sold.
What BioStack proves and what it costs you
Three things follow from a seven-person company ten months old selling six-figure contracts into frontier labs.
The market exists. This is the only company in the health vertical besides Mercor with a claimed direct lab relationship attached to a contract value, however coarse. Every other claim in the space — Protege's "majority of the Magnificent Seven," Vals's "foundation model labs," Sepal's "top AI research labs" — is an unpriced assertion.
The team size required is small. Seven people. No imaging pipeline, no scanner, no hospital data agreements visible in the description. The technical moat in this business is not headcount.
You are not first, and the first mover is invisible. BioStack does not appear in the market maps, the analyst decks or the vendor lists. That cuts both ways: it means the position is not yet crowded, and it means competitive intelligence in this vertical is unreliable by default. If a seven-person company can be selling to labs for ten months without registering anywhere, assume there are others. That is the honest posture for What the health dossier could not establish.
Its YC funding amount beyond the standard deal, its named lab customers, the term or renewal status of the six-figure contract, and whether the 17 "customers and prospects" splits closer to 2/15 or 15/2. The distinction between a customer and a prospect is doing a great deal of work in that sentence.
Sepal AI: the price nobody disclosed
Sepal was founded in 2024 in San Francisco, Y Combinator S24. Founding team: Robi Lin (ex-Turing, ex-Bain), Kat Hu (ex-Turing, ex-McKinsey) and Fedor (early engineer at Vercel and Newfront). Core team 13–15 people. It raised approximately $500K pre-seed in September 2024 from Y Combinator, Metaplanet Holdings, SID Venture Partners, Sterling Road and Team Ignite Ventures (RL List).
The product was "high-quality training data, expert-graded evaluation benchmarks, and reinforcement-learning environments for frontier LLMs," drawing on a network of 20,000+ domain experts including PhDs and professionals in finance, medicine and STEM. Its most visible public artefact was SheetBench-50, a financial-analyst-grade agent benchmark validated by finance professionals from PwC and Charles Schwab.
Mercor acquired Sepal on 6 February 2026. Orrick advised Mercor with a deal team led by Mark Seneca and John Harrison plus 24 additional attorneys (Orrick). Mercor's stated rationale: "Sepal is an exceptional team working at the intersection of human data, reinforcement learning environments, and specialized research projects," strengthening its capability in "long-horizon AI research."
The price is not disclosed anywhere. Five sources were checked and none carries consideration: the Orrick release, PitchBook's public profile stub, the Signalbase M&A entry, the founders' own LinkedIn announcement, and the RL List vendor profile.
Given ~$500K raised, a 13–15 person team, YC ownership and the acqui-hire framing of the announcement, the reasonable read is a modest, largely stock-based transaction — structurally an acqui-hire. That is an inference from structure, not a fact, and it should be labelled as such wherever it is repeated. A team of fifteen with a $500K pre-seed and a live lab relationship could equally have commanded a real number; nobody outside the deal knows.
What Mercor is actually buying
The pattern across Mercor's two acquisitions is unambiguous.
| Acquirer | Target | Date | Price | What it added |
|---|---|---|---|---|
| Mercor | Sepal AI | Feb 2026 | Not disclosed | RL environments, expert-graded evals, specialised research projects |
| Mercor | Deeptune | Jul 2026 | Not disclosed | Computer-use / enterprise environment capability, New York team |
Sources: Orrick, RL List — Mercor.
Mercor is buying environment-construction teams, not expert networks. It already has 30,000+ vetted experts and a built healthcare vertical spanning physicians, RNs, NPs, pharmacists, coders and case managers at $50–250/hr (Mercor healthcare). What it does not have enough of is people who can turn a domain into a scored environment. Sepal's 20,000-expert network was almost certainly the less interesting half of the transaction.
That is the clearest signal in the acquisition record about where value sits in this business — and it points away from supply and toward task design, exactly as The physician panels argues from the opposite direction.
BioStack is the competitor that matters and the one nobody has mapped: seven people, ten months, six-figure lab contracts, executing the thesis. Sepal is the exit comparable that does not exist, because no price was ever published — which means there is no public valuation benchmark for a clinical expert-data company, a real problem for capital planning. Only one transaction anywhere near this space has a public price, and it is Datavant/Ciox at ~$7B for a records business (Healthcare Dive) — see The real-world data brokers and What the public market pays for labour.