VaidyaAI 🧬

AI-powered multilingual medical triage assistant for Indian clinics

VaidyaAI helps patients describe symptoms, get instant triage assessments, and understand medical reports — in English, Hindi, or Kannada.

🌐 Live demo · API

The API runs on a free instance that sleeps after 15 minutes idle, so the first reply after a quiet period takes ~50 seconds. Replies after that are fast.

---

Features

• 🎙️ Voice Input — Speak symptoms in English, Hindi, or Kannada (Whisper via Groq)
• 🤖 AI Triage Agent — LangChain + Groq LLM assesses urgency and recommends specialists
• 📚 RAG Medical Knowledge — FAISS vector search over a curated symptom knowledge base
• 📄 Medical Report Analyser — Upload blood reports/PDFs, get simple explanations
• 🗣️ Text-to-Speech — Agent responses spoken aloud in patient's language
• 🏥 Doctor Dashboard — Patient queue with risk levels and full conversation history
• 🗄️ Persistent sessions — Conversations stored in Postgres (SQLite by default locally)

Triage safety model

The LLM produces a structured assessment, and a deterministic keyword check runs alongside it purely as a safety net — if it detects emergency symptoms the model may have understated, an emergency banner is prepended to the reply. It never appends a competing risk level, so a response can't show two different verdicts. A session's stored risk level only ever escalates, never downgrades.

Conversation history is rebuilt from the database on every turn rather than held in memory, so context survives a restart and stays consistent across workers.

---

Tech Stack

Layer	Technology
Frontend	React 18, React Router, React Markdown
Backend	FastAPI, Python 3.11
AI Agent	LangChain, Groq (Llama 3.3 70B)
RAG	FAISS + fastembed (MiniLM-L6-v2 via ONNX)
Voice	Groq Whisper large-v3 (STT), gTTS (TTS)
Database	PostgreSQL / SQLite, SQLAlchemy
PDF	PyMuPDF
Hosting	Vercel (frontend), Render (API)

Embeddings run through fastembed's ONNX runtime rather than sentence-transformers so the container doesn't need torch — that's the difference between a ~2 GB image and a ~250 MB one, which is what makes it deployable on a small free instance.

---

Project Structure

VaidyaAI/
├── backend/
│   ├── app/
│   │   ├── agent/          # LangChain triage agent + risk assessment
│   │   ├── api/            # FastAPI routes
│   │   ├── models/         # SQLAlchemy models, session/message tables
│   │   ├── rag/            # FAISS retrieval pipeline
│   │   ├── services/       # Voice (STT/TTS), report analyser
│   │   └── utils/          # Config
│   ├── data/
│   │   ├── medical_docs/   # Symptom knowledge base (source text)
│   │   └── faiss_index/    # Prebuilt vector index
│   ├── Dockerfile
│   └── requirements.txt
└── frontend/
    └── src/
        ├── components/     # DnaLogo, Sidebar
        ├── pages/          # PatientPage, DoctorPage
        └── services/api.js # Single API client

---

API

All routes are prefixed with /api/triage.

Method	Route	Purpose
GET	/health	Service health check
POST	/chat	Send a message, get a triage response
POST	/transcribe	Audio → text (auto-detects language)
POST	/speak	Text → speech audio
POST	/analyze-report	Upload a PDF/text report, get a plain-language explanation
GET	/sessions	All sessions with risk level and message count
GET	/sessions/{id}	One session with its full conversation
DELETE	/session/{id}	Delete a session

curl -X POST https://vaidyaai-api.onrender.com/api/triage/chat \
  -H 'Content-Type: application/json' \
  -d '{"message":"I have chest pain","session_id":"demo","language":"en"}'

---

Setup

Backend

cd backend
python -m venv venv
source venv/bin/activate      # macOS/Linux
# venv\Scripts\activate       # Windows

pip install -r requirements.txt

cp .env.example .env          # then add your GROQ_API_KEY

uvicorn app.main:app --reload

Runs at http://localhost:8000. Only GROQ_API_KEY is required — without DATABASE_URL it falls back to a local SQLite file, and tables are created on first boot.

Frontend

cd frontend
npm install
npm start

Runs at http://localhost:3000 and talks to localhost:8000 by default.

Environment variables

Variable	Required	Default
GROQ_API_KEY	yes	—
GROQ_MODEL	no	llama-3.3-70b-versatile
GROQ_FALLBACK_MODEL	no	llama-3.1-8b-instant
WHISPER_MODEL	no	whisper-large-v3
DATABASE_URL	no	local SQLite file
ALLOWED_ORIGINS	no	localhost:3000,localhost:5173
REACT_APP_API_URL	frontend, in production	http://localhost:8000/api

REACT_APP_API_URL must include the /api suffix, and Create React App bakes it in at build time — changing it requires a rebuild, not just a restart.

---

Deployment

The backend ships as a container; backend/Dockerfile works on Render, Railway or Fly without changes and reads $PORT from the host. Set GROQ_API_KEY and ALLOWED_ORIGINS (your frontend's origin) in the host's environment.

The frontend is a static CRA build — point the host at frontend/ and set REACT_APP_API_URL to the deployed API base.

---

Disclaimer

VaidyaAI provides triage guidance only. It is not a diagnosis and does not replace a qualified doctor. Every response carries this disclaimer, and emergency symptoms are flagged for immediate medical attention.