HealthVet

HealthVet is a multi-agent system that automates the vendor vetting process for U.S. health systems. When a hospital is evaluating a new clinical AI tool or EHR platform, procurement teams normally spend 2-6 weeks manually verifying certifications, checking for data breaches and lawsuits, reviewing SOC 2 reports, and cross-referencing compliance standing. HealthVet compresses that into under 90 seconds. Six specialized AI agents coordinate through Band, each owning a distinct layer of the vetting pipeline: Scout searches the web for customer references, breach history, and legal issues; Forensics analyzes certifications and flags anomalies; Compliance checks FDA, ONC, HIPAA, and SOC 2 standing; Gap maps every evidence category with severity ratings; Risk makes the final adversarial verdict and if evidence is too thin, it issues a veto and sends Scout back for a targeted re-investigation before making a final non-negotiable call; Synthesis delivers the full auditable trust report. The system is powered by AI/ML API for the reasoning-heavy agents, giving us access to gpt-4o and gpt-4o-mini through a single OpenAI-compatible endpoint, with each model tier matched to the cognitive demands of the task. The final vendor trust report is generated by Qwen2.5-72B running on Featherless AI, a deliberate choice: in a regulated industry, the compliance artifact needs to be produced by an open-weight model so hospitals and their legal teams can fully audit what generated their vendor report. Featherless makes running a 72B parameter model serverless and instant, with no GPU infrastructure required. HealthVet also includes a hospital onboarding wizard that captures each institution's specific priorities, weighted scoring across 9 categories, configurable deal-breakers, and risk appetite, so every verdict is calibrated to the hospital and not a generic rubric.