Virgil.ai: Agentic Health-Insurance Assistant

The problem

Choosing a health plan in the United States is slow and error-prone. Applicants have to check whether their medications are covered, compare plans line by line, and then wait days for underwriting. One Health, a broker agency, partnered with our team to automate that path so an applicant could be onboarded, checked and quoted in a single session.

What I built

• A conversational onboarding flow that collects the applicant's details and drives plan selection without forms to fill in by hand.
• A FastAPI service that exposes the agent to the web app over REST.
• LangChain agents on an OpenAI reasoning model, with MCP and custom tools that look up medication prices and check coverage across the candidate plans.
• An automated underwriting step that returns a decision in minutes rather than days.
• Docker images orchestrated with Kubernetes on AWS, so the service scales with a potentially large user base.

Security and compliance

Health data is regulated. Storage and transfer of AI and applicant data were designed around HIPAA requirements and general security best practice, which is what made the product usable by a licensed broker rather than a demo.

Outcome

Underwriting in minutes instead of days, medication coverage checked automatically, and onboarding automated for the broker partner. Compared with the market standard, the ease of use made Virgil.ai a viable option for real clients, not just a proof of concept.