Virgil HealthCare AI: Health-Plan Selection and Underwriting Cut From Days to Minutes Under HIPAA

Challenge

Selecting a healthcare plan is one of the most confusing tasks a US consumer faces: dozens of plans, each with its own formulary, deductibles and underwriting questions. One Health, a broker agency, collaborated with our team to remove that friction. Brokers were spending hours per applicant looking up medication prices plan by plan, and underwriting took days of back-and-forth before anyone could enrol.

Approach

• Built Virgil as an agentic AI system on LangChain and FastAPI, with an OpenAI reasoning model as the backbone and custom tools exposed through MCP, so each capability is a scoped, testable action rather than a single prompt.
• Automated the hardest manual task: an efficient medication price lookup that checks whether an applicant's prescriptions are covered by each candidate plan, and at what cost.
• Automated underwriting so the applicant's answers and documents are evaluated in minutes, with a human broker reviewing edge cases rather than every file.
• Designed the whole system for HIPAA: encrypted storage and transfer, least-privilege access, and no protected health information in prompts beyond what the task needs.
• Deployed on AWS with Docker and Kubernetes so onboarding scales with demand.

Results

• Underwriting reduced from days to minutes.
• Medication coverage and price lookup automated across plans, instead of hours of manual checking per applicant.
• A large potential user base and far easier experience than market standards made the product a viable option for the broker's clients.

AI-readiness impact

This is the governance pattern we bring to HubSpot agent rollouts: one capability at a time, each exposed as a scoped tool; humans in the loop on the decisions that matter; and data handling designed for the regulation before the first user, not after. The same discipline applies to a Breeze Customer Agent with authenticated actions on sensitive customer data.