Healthcare Explorer — MCP & OpenAI
An interactive healthcare-data demonstration for exploring generated patient profiles, risk signals, readmission indicators, and database structure. A FastAPI browser application exposes read-only SQLite data through four MCP tools, with a separately access-controlled OpenAI assistant.
Public explorer. AI chat requires a private access code.
The problem
An interactive healthcare-data demonstration for exploring generated patient profiles, risk signals, readmission indicators, and database structure. A FastAPI browser application exposes read-only SQLite data through four MCP tools, with a separately access-controlled OpenAI assistant.
My contribution
I integrated the MCP tools and bounded OpenAI tool-calling client with a responsive browser explorer, generated demonstration database, read-only SQL controls, and Vercel delivery.
Product features
- Search and inspect 10 generated patient profiles with linked demonstration records.
- Explore rule-based risk signals and historical readmission indicators.
- Inspect five related SQLite tables in the schema view.
- Four MCP tools expose schema discovery, bounded SQL queries, risk summaries, and readmission insights.
- An access-code-protected AI assistant discovers and calls the read-only MCP tools.
Engineering work
- Structured data access: Expose schema, bounded read-only SQL, and demonstration summaries through four discoverable MCP tools.
- Browser exploration: Connect patient search, profile details, and schema views to a separately generated synthetic SQLite database.
- Controlled AI access: Bound model tool loops and generated SQL, require an AI access code, and separate illustrative heuristics from clinical claims.
Outcome & scope
Makes 10 synthetic profiles and five related tables explorable through a browser and four structured tools. Live checks verified search, profile details, and schema browsing; no clinical accuracy, patient outcomes, or real healthcare adoption are claimed.
Generated-data demonstration · Read-only tools · Risk heuristics are not clinical predictions