Conviction

Conviction, which operates the product Parachute, builds governance infrastructure for clinical artificial intelligence. Its platform is aimed at hospitals and health systems that need to evaluate, approve, deploy and continuously monitor AI tools — including electronic health record plug-ins such as Epic extensions, third-party vendor models, and models developed in-house, such as imaging algorithms. The company positions the product around two stated problems: patient-safety risk from AI errors and the cost of slow or stalled AI pilot programs.

The product is organized into four phases. Discovery and prioritization uses AI agents to map a hospital's clinical gaps, goals and budget to available vendors, while a review engine flags compliance issues in vendor documentation. Evaluation and approval applies automated scoring and stress testing of accuracy, safety and bias in place of manually written test scripts. Deployment and monitoring provides real-time observability of performance, drift and bias, with alerts when a metric falls below a threshold. An audit-and-prove phase maintains a documentation trail of approvals, tests, changes and runtime metrics intended for regulators and litigation. Parachute also operates a Vendor Marketplace that ranks AI vendors against hospital-specific needs by fit, risk and impact, and states that the platform integrates with a hospital's existing stack.

The company frames its market around regulatory requirements it cites, including HTI-1, the Colorado AI Act, California SB 3030 and the White House AI Action Plan, alongside its claim that more than 2,000 clinical AI tools have entered the market and that lean hospital IT teams cannot vet and monitor each model.

Founding story

Co-founders Aria Vikram (Founder/CEO) and Tony Yamin (Founder/CTO) say they conceived Parachute while deploying AI at Columbia University Irving Medical Center, where they repeatedly encountered obstacles to hospital AI adoption: a proliferation of clinical AI tools, new rules requiring auditable evidence of safety, equity and efficacy, small IT teams unable to vet and monitor every model, and pilots that failed to reach production for lack of deployment infrastructure. Vikram was previously an NLP researcher at Columbia and deployed AI chatbots for doctors at CUIMC; Yamin previously built machine-learning infrastructure for the U.S. Department of Defense at Booz Allen.

Business model

Software sold to hospitals and health systems for AI governance, combined with a two-sided Vendor Marketplace in which AI vendors apply to be listed and matched to hospital needs; a launch offer of 50% off a marketplace listing was extended to Y Combinator AI companies.

Traction

Columbia University Irving Medical Center uses Parachute to track live models in real time, according to the company's Y Combinator launch post. The company was named in a Y Combinator-linked list of promising startups from the Summer 2025 batch.

Latest developments

Parachute was publicly launched during the Y Combinator Summer 2025 batch, and the company published content on clinical AI governance, the White House AI Action Plan and its Vendor Marketplace. A September 4, 2025 article named it among promising startups from the batch.