VA Assist

A clinician has minutes with a patient and decades of records in VistA. Finding the three facts that matter means clicking through screens built long before anyone imagined AI. The data was there. Getting to it in time was the problem.

We built VA Assist, a first-of-its-kind generative AI product that lives inside the VA's Clinical Decision Support platform. We used the Model Context Protocol to make VistA data AI-native, so agents can read the patient record through the same RPCs the VA already trusts. Several agents gather clinically relevant data and reviews in parallel, then hand the clinician a short, sourced summary in plain language. PII is filtered automatically, and sign-in rides on the VA's existing security.

• MCP server that makes VistA data AI-native
• Agents that gather clinical data and reviews in parallel
• Natural language interface with automatic PII filtering
• Security and authentication through existing VA systems

VA Assist is the first AI pilot the VA is deploying to improve clinical decision times. It lowers cognitive load for clinicians and gives patients a better visit, with full security and compliance. We also convinced the VA to open-source it, so other teams can adopt it as a standard instead of rebuilding it.