RediSense

Problem

Capstone for the Atomcamp × NUST Data Science & AI Bootcamp: a clinical copilot that brings medical report analysis, symptom triage and evidence lookup into one workflow for healthcare teams.

What I built

• Medical report upload and analysis, symptom triage and a per-patient timeline, each persisted and written as timeline events.
• Evidence search over a locally ingested dataset using Postgres with pgvector, and a copilot chat with database-backed conversations and optional patient context.
• Risk-prioritised reports dashboard and severity-based triage views.
• FastAPI backend with SQLAlchemy and Alembic migrations, Next.js App Router frontend, and a demo mode that falls back to curated mock data when the backend is unavailable. Shipped with CI.
• Keeps working without an OpenAI key by using conservative heuristic fallbacks for analysis, triage, copilot and embeddings.

Stack