Healthcare AI has crossed an important threshold. The question is no longer whether models can contribute meaningfully to clinical and scientific work, but rather if they can do so safely and reliably.
That distinction matters more with every model release. AI is moving rapidly from administrative tasks into increasingly complex applications across healthcare and life sciences, from clinical reasoning and documentation to drug discovery and research. Yet even models that perform exceptionally well on general benchmarks can struggle with the nuance, judgment and edge cases that clinicians and scientists navigate in the real world. In healthcare, failures aren't bugs to patch in the next version. They carry real consequences for patients, providers and the drug development process.
Addressing these challenges requires training and evaluating AI against standards grounded in real clinical and scientific practice. That is precisely what Alterra Labs is building, and why Oak is proud to lead the company's $12.5 million seed round.
Alterra builds the reinforcement learning (RL) environments and evaluations that train AI models to perform complex clinical and scientific work safely and reliably. Its environments simulate the systems where that work takes place, across the electronic health record, PACS, payer portals, clinical trial software, consumer applications, and scientific research tools. By evaluating models as they navigate complex, multi-step tasks within these settings, Alterra’s researchers identify where performance breaks down, analyze why failures occur, and generate the feedback needed to improve models before they are deployed within real-world environments.
The quality of these environments depends as much on clinical and scientific expertise as it does on technical sophistication. Alterra’s in-house team of US-trained and licensed physicians, scientists and ML researchers work together alongside a network of partners to design tasks and evaluation criteria. Their involvement ensures that training signals capture not only whether a model reaches the correct answer, but whether the reasoning and decisions behind it reflect the standards that professionals apply in their own work.
This combination of domain expertise and technical rigor allows Alterra to create value for both those developing advanced AI and those responsible for deploying it. For frontier labs and applied AI companies, Alterra identifies where models fail on specialized healthcare and scientific workflows, providing the training environments and feedback needed to improve performance. For health systems and life sciences organizations, it evaluates how AI performs within their specific clinical and operational contexts, helping inform deployment decisions, appropriate safeguards, and ongoing monitoring.
It also sets Alterra apart in a growing field. Most players in RL environments and evaluation are horizontal, meaning they are native to the AI lab world and treat healthcare as one vertical among many. Alterra is built the other way around, with clinical depth at its core and the technical rigor to partner with the most sophisticated labs in the world.
The broader direction of AI development makes this work especially timely. As gains from pre-training plateau, reinforcement learning has emerged as one of the most efficient levers for improving model capabilities. AI labs already spend roughly $7 billion annually on training data, a figure on pace to approach $100 billion by 2030. As this investment grows, labs are increasingly turning to specialized partners for complex, domain-specific environments where evaluation and reward design require substantial depth and expertise. Healthcare sits squarely in that category, yet there is no healthcare-first leader in RL environments today. We see a clear opening for Alterra to define the category.
Ultimately, our conviction starts with the team. Founder and CEO Akash Kapoor started the company while training as a medical student at Columbia University, after having worked as a machine learning researcher at the National Institutes of Health. His experience across both fields gave him firsthand insight into the potential of advanced AI, as well as the challenges of ensuring that these systems meet the standards required in clinical and scientific settings. He is joined by Dr. Morgan Jeffries, Chief Clinical Officer, a neurologist and former Medical Director for AI at Geisinger, whose experience integrating AI into a large health system informs Alterra's approach to clinical rigor. Morgan now leads the company's clinical fellowship program, bringing together the specialized knowledge needed to evaluate and improve these systems responsibly.
Our partnership with the team extends well beyond capital. As the company’s ambitions expanded, its original name no longer reflected the breadth of its work. Beginning this summer, Oak’s platform team worked closely with Alterra to sharpen its mission and narrative, ultimately developing a new name, visual identity and website. The result, Alterra Labs, reflects a company whose mission spans the entire healthcare and life sciences ecosystem, with an identity that speaks to both the frontier researchers advancing AI and the clinicians and scientists working to translate its capabilities into practice.
With this funding, Alterra will accelerate its research, scale its in-house team of clinicians, scientists and researchers, expand its expert-in-residence network across specialties and deepen partnerships with the organizations developing and deploying advanced AI.
AI has extraordinary potential to transform healthcare and life sciences, but realizing that potential will depend on more than the capabilities of the models themselves. It will require the infrastructure to evaluate and improve how these systems perform in the settings where their reliability matters most and to continuously improve them to deliver the greatest positive impact on human health. Alterra is building that infrastructure, and we are proud to partner with Akash and the entire team as they take on one of the most consequential challenges in AI.
