Tl;Dr;

We build realistic healthcare simulation environments where AI labs can train and improve their models before deploying them in the real world. Our environments use real-world clinical data to help models practice solving the hardest medical problems.

The problem

Healthcare AI is learning from a cleaned-up version of medicine that barely exists. Real patients are messy. Their records are incomplete. Their labs are late. Their notes conflict. Their outcomes unfold over months, not in one prompt.

Doctors do not just answer questions. They make decisions with imperfect information, watch what happens, and adjust. That is the loop AI needs to perfect.

Today, AI labs have medical data. What they do not have are real clinical environments where models can practice, fail, get feedback, and improve.

What BioStack does

• We turn real patient records into training environments. Not clean textbook cases. Real longitudinal data: notes, labs, imaging, meds, vitals, diagnoses, follow-ups, and outcomes stitched into cases models will actually have to work through.
• We build the tasks models train on. Diagnose this. Order the next test. Adjust the medication. Flag the risk. Predict what happens next. The model makes a clinical move, then gets scored against outcomes, chain-of-thought reasoning, guidelines, and clinician review.
• We give AI labs the full post-training loop for healthcare. Data, evals, reward functions, benchmarks, and simulation environments, so models can practice real clinical work instead of just memorising medical exams.

https://drive.google.com/file/d/1NvcLS2usN_bl4GE_tW7QVN5Ia-MGl8M6/view

Backstory

We started the company in Oct 2025 and burned through two pivots before we found the thing that pulled us in.

First, we tried building multi-agent systems for biology research. It sounded right on paper. In practice, biology workflows are messy, fragile, and full of assumptions that have to be tuned by hand. One wrong parameter compounds downstream and suddenly nobody trusts the output. Scientists also were not ready to hand over real work to agents that still felt too brittle. We killed it after two months.

Then we tried going deeper: generate preclinical drug discovery data ourselves, at scale, through our own lab facility. Again, it sounded ambitious. Again, it broke on contact with reality. It was capex-heavy, slow, operationally painful, and not truly venture-scalable. Worse, our edge was basically cheaper offshore execution. That is not a moat. That is a race to the bottom. We killed that too.

The third idea found us.

The best companies are born in the customer’s office. For us, it happened on a sales call.

We were pitching drug discovery data to one of the largest human data companies in the world. Halfway through, the conversation shifted. They did not want another generic data vendor. They asked if we could deliver high-quality clinical data for a six-figure contract.

That call changed the company.

We realised the real bottleneck was not just raw data. AI labs need the full post-training stack for healthcare: clinical data, simulation environments, evals, reward functions, and verifiable tasks that make models better at real medical work.

So we followed the pull.

Today, BioStack is building simulation environments for healthcare AI. We have expanded from one customer conversation into 17 customers and prospects, including top AI labs, and we are seeing demand across data, evals, and RL environments.

We did not arrive here through a brainstorm but got dragged here by the market.

Asks

• Do you know anyone working on a AI models or agents advancing current capabilities through multimodal applications, nuanced reasoning, or domain-specific expertise?
• Do you know of companies that require raw medical data? We supply longitudinal high quality scans, EHR, and notes and we’re happy to send over samples.
• Have referrals for RL engineers / researchers? We’re always looking to hire RL engineers who specialise in fine-tuning methods such as SFT, PEFT, and RL.
• Spread the word - send us ideas for data types that you think are valuable and spread the word!

Feel free to reach out at: sanat@getbiostack.com / parth@getbiostack.com