Alterra Labs Raises $12.5M Seed for Medical AI Evaluations

Alterra Labs considers an AI model to have failed when it reaches a correct answer for the wrong reasons. That grading rule puts clinical judgment inside the product the New York company is financing with its $12.5M seed round, announced November 10, 2026, led by Oak HC/FT with Flare Capital Partners, Company Ventures, Tau Ventures and angels participating.

Alterra Labs builds simulated healthcare and life-science workflows in which models can be evaluated and improved. Founder and CEO Akash Kapoor is building around an unusually demanding customer requirement: the people buying medical AI need an explanation of how it behaves while doing the work. The funding supports more research, a larger clinical and scientific team, and deeper partnerships with organizations developing advanced AI.

Clinical judgment becomes part of the training infrastructure

On its About page, Alterra describes evaluating reasoning alongside actions and outcomes. That choice gives the clinician a role in defining success before a model receives feedback. A technically completed task can still contain a decision the evaluator would reject, which makes the standard being applied part of what the customer is buying.

The company's website describes environments spanning electronic health records, imaging, patient-facing tools, clinical trials, payer workflows and scientific research. In an electronic health record, the work can involve a patient's history across visits; in payer workflows, it can involve appeals, authorization and coding. Those descriptions establish the range of tasks Alterra wants to make evaluable, without claiming that one score can certify every use of a model.

The operating implication is substantial. Someone has to decide what the task should ask, which information the model should consider, and what would make its actions acceptable. A useful evaluation therefore carries a specialist's understanding of the job into a format that a model developer can inspect and use. That relationship creates recurring work as the underlying software, medical knowledge and model behavior change.

For the customer, the explanation of a failure can be as useful as the score. It gives a researcher a specific behavior to work on and a buyer a concrete issue to discuss with a vendor, rather than leaving both sides to argue about an impressive aggregate result.

The seed round finances people who can define the work

Oak HC/FT's investment account places clinical expertise at the center of its rationale. Akash Kapoor previously conducted machine learning research at the National Institutes of Health and trained at Columbia Medical School. Dr. Morgan Jeffries, Alterra's Chief Clinical Officer, is a neurologist and former Medical Director for AI at Geisinger who leads the company's clinical fellowship program.

Those backgrounds matter to the construction of the product. Evaluating a workflow requires choices about what a professional should know and do in that setting, alongside the engineering needed to observe a model's attempt. The clinical fellowship brings specialist knowledge into the work of forming those judgments, giving the company a way to develop expertise together with its evaluation infrastructure.

The founder's announcement identifies Oak as lead investor and names Flare Capital Partners, Company Ventures and Tau Ventures as participants, alongside unnamed angels. It states that the money will accelerate research, expand the in-house team and deepen partnerships. The official sources checked do not specify the dollar denomination, valuation or cumulative funding total, so the financing is reported using the company's own $12.5M figure.

Hiring clinicians and scientists is therefore closely connected to product capacity. More environments require more choices about realistic tasks and meaningful assessment. For investors, supporting that work means financing the people who make an evaluation worth purchasing as well as the software that delivers it.

Model developers and healthcare buyers need different answers

Flare Capital Partners describes a business serving frontier labs, applied AI companies and organizations evaluating tools for deployment. It reports that Alterra has converted early engagements into paid partnerships with a handful of frontier labs. The investor does not identify those customers or disclose contract values, leaving the scale of the commercial relationships outside the public record.

A model developer wants to locate a failure and produce feedback that helps improve a subsequent attempt. A health-system buyer faces a deployment decision involving a particular workflow, its safeguards and the people who will use the output. An evaluation may inform both conversations, but each customer has a different reason to pay attention to the result.

That creates a useful commercial connection for Alterra. Work performed to understand a model can inform its improvement, while the same discipline can help an organization decide where the model belongs. The relationship depends on the evaluation remaining legible to the specialist making the decision. An elaborate simulation that nobody can interpret would leave the buyer with another demonstration to admire and another decision still waiting.

As Alterra expands its clinical and research capacity, the work keeps returning to the grading rule. A physician defining an acceptable action is helping determine what a researcher will try to improve next, and what a healthcare organization will ask to see before bringing that model into its own workflow.