Patient-Facing Clinical Trial Chatbot with Interactive Map

Cancer patients navigating clinical trial databases face complex search interfaces with dozens of technical filters, making it difficult to find relevant trials without medical expertise. This project builds a patient-facing clinical trial chatbot with an interactive map that removes this barrier through natural language. The system features a three-panel interface: a chat panel powered by an LLM agent that elicits patient information and matches it against the Cancer Trials Canada database, an interactive map that displays geocoded trial site locations and a summary panel that shows trial details when a pin is clicked Built using Chainlit server mounted inside FastAPI, React application with Mapbox GL for the interactive map. Trials are stored in PostgreSQL with pgvector for semantic and full-text search. An open-source LLM served via Ollama and Pydantic-AI. Multi-layered guardrails prevent off-topic or harmful responses, RAGAS for evaluation and Langfuse for observability and tracing throughout development.