Skip to content

Altira Labs raises $12.5 million to train AI on simulation and clinical decision-making

Share
Altira Labs raises $12.5 million to train AI on simulation and clinical decision-making

Altira Labs, a startup in New York, raised $12.5 million in a seed round led by Oak HC/FT, with participation from Flare Capital Partners and Company Ventures, to build digital simulation environments specialized in testing and training medical AI models. This round marks an explicit shift in AI lab priorities, focusing on giving models proper clinical behavior rather than merely stacking theoretical medical data.

The company uses a mechanism that places models inside virtual clinical tasks to identify their points of failure, then turns those failures into intensive training exercises. Physicians and scientists help design these cases and evaluate model choices, such as determining which lab tests to order for a virtual patient, while verifying the medical appropriateness of those decisions. Altira Labs converts these clinical evaluations into registration metrics and rewards used in reinforcement learning, encouraging models to repeat sound decisions and avoid practices that experts label unsafe.

Founder and CEO Akash Kapoor says current models can learn to use software, but they stumble when faced with clinical reasoning,as the models do not recognize when to refrain from a medical procedure and lack how to manage uncertainty.This description matches the findings of Stanford University researchers on the “Physician Bench” and “Health Admin Bench” standards for testing intelligent agents, where even the best models failed to complete half of the assigned tasks, some models noted the correct procedure in medical notes without actually executing it in the electronic health record, while other models lost vital data when switching between software systems.

Vijay Chandramouli, a partner at Oak HC/FT, sees health-AI companies moving beyond abstract data research toward embedding actual workflow execution by simulating electronic health record systems, insurance gateways and billing. Building these environments requires expertise that blends deep medical practice with a understanding of model behavior, a path embodied by Kapoor, who studied machine learning before studying medicine at Columbia University and conducting research at the National Institutes of Health, before leaving medical school in his final year to devote himself to the company that launched in 2025 as “ChartR.”

The founder’s concern for healthcare safety grew from a personal experience during his university studies, when he found in his medical file a two-year-old radiology report that had been ignored, warning of a tumor in his leg and recommending urgent referral, later revealed to be a malignant bone tumor that required surgery. Today Altira Labs, through environments that also include medical imaging, clinical trials and tele-health, focuses on cases that do not match standard textbook descriptions, where models have abundant knowledge but lack the wisdom of judgment. The company generates revenue by selling training environments and post-training services, drawing on a client base led by advanced AI labs, hospitals and medical societies.

This shift carries a direct professional significance for the health sector in the Gulf, Egypt and the Levantas hospitals engage in a broad race to digitize records and adopt intelligent diagnostic technologies. The challenge is no longer purchasing generative model licenses or training them on general medical text, but attracting and building personnel who understand the specifics of clinical workflows and can audit model behavior within real-world operating environments. Regional health institutions now need physicians and health-informatics specialists who can test software agents when integrating them into care and billing systems, ensuring data flow stability and decision safety. The decisive standard becomes linked to staff ability to train systems and hold them accountable when a patient’s condition does not fit textbook templates.

Don't miss the next story

Subscribe for updates