Swasthya AI

A two-sided clinical intelligence platform for continuous patient monitoring and doctor decision support — built for the Indian healthcare context.

What It Is

Swasthya AI bridges the gap between patients managing chronic conditions and the care they need between visits. Patients log symptoms, take daily AI check-ins, and track medications on a mobile app. Doctors receive AI-generated morning briefings, scan patient QR codes for instant clinical context, and query an AI assistant grounded in each patient's actual health data.

Stack

Mobile App — React Native 0.81, Expo 54, TypeScript, Expo Router, Zustand, Supabase JS, Apple HealthKit / Google Health Connect

Doctor Dashboard — React 19, TypeScript, Vite 6, Tailwind CSS v4, Recharts, html5-qrcode, Supabase JS

Backend — FastAPI (Python 3.11), Groq LLaMA 3.3 70B, LangChain + FAISS (RAG over medical guidelines), scikit-learn (Isolation Forest + Linear Regression), OpenFDA API, Supabase, N8N automation, Docker

Core Features

• AI conversational onboarding — no forms; multi-turn dialogue collects full health profile
• Daily adaptive check-ins — 2–3 personalised questions generated from conditions, symptoms, and last 24h of wearable data
• Body heatmap — symptom history mapped to anatomical zones with recency decay
• Risk score — deterministic base score + RAG-adjusted by LLM (±15 max, guideline citation required); always shown as a range with confidence band
• Drug conflict detection — synchronous OpenFDA check blocks medicine save until complete
• Wearable anomaly detection — SQI filter → per-patient Isolation Forest + statistical threshold
• Deterministic escalation matrix — 8 critical single-symptom overrides + two-signal combinations; LLM writes the message, never makes the trigger decision
• Jan Aushadhi PDF — printable generic medicine reference mapped from branded prescription
• Government scheme matching — PM-JAY, RAN, PMJAP, and state schemes matched to patient profile
• Family health system — privacy-filtered shared view; sensitive conditions excluded at database query level
• Morning briefing — AI-generated nightly; only flagged patients shown, sorted by urgency
• QR scan entry — individual or family QR → full clinical context in under 3 seconds
• Doctor AI assistant — free-text clinical questions answered from patient data only, with RAG source visibility
• Async Q&A loop — doctor queues questions → patient answers in next check-in → doctor notified
• Appointments section — clinical priority queue based on health flags, not a calendar
• Recovery Counter — objective professional metric; patients recovered under a doctor's care

Setup

# Backend
cd backend && pip install -r requirements.txt
# Add backend/.env (GROQ_API_KEY, SUPABASE_URL, SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY)
uvicorn main:app --reload

# Mobile App
cd app && npm install
# Add app/.env (EXPO_PUBLIC_SUPABASE_URL, EXPO_PUBLIC_SUPABASE_ANON_KEY, EXPO_PUBLIC_API_BASE_URL)
npx expo start

# Doctor Dashboard
cd web && npm install
# Add web/.env (VITE_SUPABASE_URL, VITE_SUPABASE_ANON_KEY, VITE_API_BASE_URL)
npm run dev

Place medical guideline PDFs in backend/rag/guidelines/ — FAISS index builds automatically on first startup.

Safety Design

All AI outputs are risk indicators, not diagnoses. Escalation logic is a deterministic Python matrix — the LLM cannot override it. Drug conflict detection is a blocking gate. RAG adjustments to risk scores are bounded and require guideline citation. Sensitive conditions (HIV, mental health, cancer, STDs) are filtered at the database query level.

License

MIT