GitHub - Swap-24/MedPathAI: AI-Powered Healthcare Navigator & Cost Estimator for India. Built for TenzorX National Hackathon

```
    ███╗░░░███╗███████╗██████╗░██████╗░░█████╗░████████╗██╗░░██╗
    ████╗░████║██╔════╝██╔══██╗██╔══██╗██╔══██╗╚══██╔══╝██║░░██║
    ██╔████╔██║█████╗░░██║░░██║██████╔╝███████║░░░██║░░░███████║
    ██║╚██╔╝██║██╔══╝░░██║░░██║██╔═══╝░██╔══██║░░░██║░░░██╔══██║
    ██║░╚═╝░██║███████╗██████╔╝██║░░░░░██║░░██║░░░██║░░░██║░░██║
    ╚═╝░░░░░╚═╝╚══════╝╚═════╝░╚═╝░░░░░╚═╝░░╚═╝░░░╚═╝░░░╚═╝░░╚═╝
```

AI-Powered Care Navigation & Healthcare Financing for India

From first symptom to financing approval — guided by AI, every step of the way.

What is MedPath AI?

MedPath AI is a full-stack conversational healthcare navigation platform built for the Indian market. Describe your symptoms in plain language — the AI figures out what procedure you need, finds the right hospitals in your city, shows a personalised cost breakdown adjusted for your medical history, and walks you through applying for a Poonawalla Fincorp medical loan. All in one chat.

No phone calls. No clinic hopping. No guesswork.

Features

Conversational clinical intake — Gemini 2.5 Flash extracts procedure, city, urgency, and ICD-10 codes from plain language
Emergency bypass — detected at ≥85% confidence, skips clarification entirely and surfaces hospitals immediately with a red alert banner
Intelligent hospital ranking — scored by cost match, quality rating, NABH/JCI accreditation, wait time, and GPS distance
Personalised cost breakdowns — comorbidity multipliers (diabetes, cardiac, kidney, etc.) applied on top of procedure base costs
PFL loan pre-qualification — instant GREEN / YELLOW / RED eligibility decision based on CIBIL, income, FOIR, and loan multiple
Gemini document extraction — upload your salary slip or CIBIL report and the AI fills in your financial profile automatically
Real-time loan tracking — patient app polls loan status every 3 seconds; sees the officer's decision the moment it's made
5-step registration wizard — health profile, comorbidities, insurance, financials, and emergency contact
Spectator-safe officer dashboard — completely separate Vite build for PFL officers; ships zero patient code

Coming Soon

AI second opinion — challenge a hospital recommendation with a follow-up question
User profiles — persistent health history and past consultation summaries
Real PFL API — live Poonawalla Fincorp loan disbursement integration

Tech Stack

┌──────────────────────────────────────────────────────────┐
│                React 19 + Vite (Frontend)                │
│        React Router 7 · Tailwind v4 · Zustand 5          │
└─────────────────────┬────────────────────────────────────┘
                      │ HTTP + REST
         ┌────────────┴────────────┐
         │                         │
┌────────▼────────┐      ┌─────────▼────────────┐
│  FastAPI 0.115  │      │  Supabase (Postgres) │
│  LangGraph      │◀────▶│  + Storage Bucket    │
│  Pipeline       │      │  9 tables            │
└────────┬────────┘      └──────────────────────┘
         │
    ┌────┴──────────────────────────┐
    ▼            ▼                  ▼
 INTENT       PROVIDER           COST + RESPONSE
 Gemini       Pure Python        Formula Engine
 2.5 Flash    Hospital           + Gemini
 Clinical     Scoring            2.5 Flash
 Intake       Engine

Layer	Technology
Frontend	React 19, Vite 8, Tailwind CSS v4, React Router 7
State	Zustand 5, custom hooks (useChat, useLoan, useDocuments)
Backend	FastAPI 0.115, Uvicorn 0.30, Pydantic v2
AI Pipeline	LangGraph 0.2.28, Gemini 2.5 Flash
Clinical Intake	`gemini-2.5-flash` (intent extraction, ICD-10 mapping)
Relevance + Scoring	Pure Python, Pandas, Haversine distance
Document Extraction	`gemini-2.5-flash` (PDF/image → structured JSON)
Database	Supabase Postgres + Supabase Storage

Getting Started

Prerequisites

Node.js 20+
Python 3.11+
A Supabase project (free tier works)
A Google Gemini API key (Gemini 2.5 Flash access required)

1. Clone the repo

git clone https://github.com/yourusername/MedPathAI.git
cd MedPathAI

2. Backend — Python environment

python -m venv venv
source venv/bin/activate          # Windows: venv\Scripts\activate
pip install -r requirements.txt

3. Backend — environment variables

Create a .env file inside backend/ (or copy from .env.example):

SUPABASE_URL=https://<your-project>.supabase.co
SUPABASE_KEY=<your-supabase-anon-or-service-key>
GEMINI_API_KEY=<your-gemini-api-key>
DOCUMENT_BUCKET=medical-documents
DOCUMENT_EXTRACTION_ENABLED=false

# JWT Authentication (optional, for production)
JWT_SECRET=<your-secret-key>
JWT_TTL_SECONDS=86400
PFL_OFFICER_API_KEY=<optional-api-key-for-officer-dashboard>

4. Frontend — environment variables

Create a .env.local file inside medpathai/ (or copy from .env.example):

VITE_API_BASE_URL=http://localhost:8000
VITE_SUPABASE_URL=https://<your-project>.supabase.co
VITE_SUPABASE_ANON_KEY=<your-anon-key>

5. Run the backend

python ./backend/main.py   # → http://localhost:8000

6. Frontend

cd medpathai
npm install
npm run dev         # → http://localhost:5173 (patient app)
npm run dev:pfl     # → http://localhost:5174 (PFL officer dashboard)

7. Verify everything is running

http://localhost:5173 → Patient chat app
http://localhost:5174 → PFL officer dashboard
http://localhost:8000 → {"status": "MedPath AI is running :)"}
http://localhost:8000/docs → FastAPI Swagger UI (test /api/chat here)

On first run the backend loads hospital, procedure, and city data from Supabase into memory. Seed CSVs are in backend/data/.

Build & Deployment

Build for production

# Patient app
cd medpathai && npm run build       # → dist/

# Officer dashboard
cd medpathai && npm run build:pfl   # → dist-pfl/

# Backend is production-ready via Uvicorn
python -m uvicorn backend.main:app --host 0.0.0.0 --port 8000 --workers 4

Deploy to Vercel (frontend)

vercel deploy

Uses vercel.json config for SPA routing.

Deploy to Railway (backend)

railway link
railway deploy

Uses railway.toml config for Python FastAPI deployment.

API Endpoints

Chat & Recommendations

POST /api/chat — Send message, get LangGraph response (intent→provider→cost→response)
GET /api/hospitals — List hospitals for a city
GET /api/procedures — List available procedures

Registration & Profile

POST /api/register — Create user profile with health details
GET /api/profile/{user_id} — Retrieve user profile
PUT /api/profile/{user_id} — Update profile info

Loan & Finance

POST /api/loan — Submit PFL loan application (runs eligibility check)
GET /api/loan/{loan_id} — Get loan status
PUT /api/loan/{loan_id}/approve — Officer approves/rejects loan

Document Processing

POST /api/extract-document — Upload PDF/image, extract via Gemini
GET /api/documents/{user_id} — List user documents

Security & Auth

POST /api/auth/login — Phone/email authentication
POST /api/auth/verify — OTP verification
POST /api/auth/logout — Clear session

Security Features

The security.py module provides:

JWT Authentication — Stateless session tokens with configurable TTL (default: 24 hours)
API Key Validation — PFL officer dashboard is protected by PFL_OFFICER_API_KEY
HMAC Signatures — Verify webhook authenticity from Poonawalla Fincorp
CORS — Restricted to frontend domains (http://localhost:5173 in dev)

Key functions:

create_jwt(user_id) — Generate token
verify_jwt(token) — Validate + decode
verify_officer_api_key(api_key) — Check officer access

How It Works

Register with your health profile, comorbidities, insurance details, and financials
Upload financial documents — salary slip, ITR, CIBIL report (Gemini extraction has been developed but not yet implemented)
Open the chat and describe your symptoms or the procedure you need
The AI asks up to 3 targeted clarifying questions — then commits to a recommendation
Get a ranked list of hospitals in your city with personalised cost breakdowns
Select a hospital and apply for a Poonawalla Fincorp medical loan in one click
The AI runs an instant eligibility check and submits your application
Track your loan status live — the PFL officer's decision appears in seconds

ML Scoring Pipeline

Every argument runs through four stages in the LangGraph pipeline:

Node	Model / Engine	Role
Intent	Gemini 2.5 Flash	Extracts procedure, city, budget, urgency, emergency flag, ICD-10 hint
Provider	Pure Python + Pandas	Scores hospitals by cost, quality, accreditation, wait time, GPS distance
Cost	Formula Engine	Personalised breakdown with comorbidity multipliers + PFL loan options
Response	Gemini 2.5 Flash	Generates warm, clinically-phrased summary or emergency alert

Comorbidity multipliers: Kidney Disease +22% · Cardiac +18% · Diabetes +12% · Obesity +10% · Hypertension +8% · Asthma +6%

PFL eligibility tiers: CIBIL ≥700, income ≥₹15k/month, FOIR ≤50%, loan ≤10× income → GREEN (instant approval) · soft flags → YELLOW · hard failure → RED (alternatives suggested)

Project Structure

MedPathAI/
├── backend/                        # FastAPI + LangGraph server
│   ├── main.py                     # All HTTP endpoints + CORS + error handling
│   ├── graph.py                    # LangGraph StateGraph (intent→provider→cost→response)
│   ├── loan_engine.py              # PFL credit policy, CIBIL checks, EMI calculations
│   ├── security.py                 # JWT auth, API key validation, signature verification
│   ├── db.py                       # Supabase client, DB helpers, ORM mappings
│   ├── data_loader.py              # Hospital search engine, cost formulas, city lookup
│   ├── test_intent.py              # Manual testing of intent extraction node
│   ├── nodes/
│   │   ├── intent.py               # Gemini clinical intake node (ICD-10 extraction)
│   │   ├── provider.py             # Hospital scoring engine (pure Python + Pandas)
│   │   ├── cost.py                 # Cost breakdown, comorbidity multipliers, loan options
│   │   └── response.py             # Gemini response generation, emergency alerts
│   └── data/                       # Seed CSVs for bootstrap
│       ├── hospitals_rows.csv      # 500+ hospital records
│       ├── procedures_rows.csv     # ICD-10 procedure mappings
│       └── cities_rows.csv         # City location data
│
├── medpathai/                      # React 19 + Vite frontend
│   ├── src/
│   │   ├── App.jsx                 # Main app router (Login → Registration → Chats)
│   │   ├── main.jsx                # Vite entry point (patient app)
│   │   ├── screens/
│   │   │   ├── Chats/              # Main chat interface
│   │   │   │   ├── index.jsx       # Chat container + message management
│   │   │   │   ├── ChatInput.jsx   # Message input + voice support
│   │   │   │   ├── MessageList.jsx # Scrollable message history
│   │   │   │   ├── MessageBubble.jsx  # Styled chat bubbles
│   │   │   │   ├── HospitalCard.jsx   # Hospital recommendation cards
│   │   │   │   ├── CostBreakdown.jsx  # Cost itemization + comorbidity display
│   │   │   │   ├── EligibilityResult.jsx  # PFL loan status (GREEN/YELLOW/RED)
│   │   │   │   ├── PFLLoanPanel.jsx    # Loan application flow
│   │   │   │   ├── ProviderModePanel.jsx  # Hospital provider mode UI
│   │   │   │   └── PossibleCauses.jsx    # Diagnostic suggestions
│   │   │   ├── Documents/          # Document upload + Gemini extraction
│   │   │   │   ├── index.jsx       # Main documents screen
│   │   │   │   ├── UploadDropzone.jsx  # Drag-drop PDF/image uploader
│   │   │   │   ├── DocumentCard.jsx    # Uploaded document preview
│   │   │   │   └── ExtractionPreview.jsx  # Extracted JSON viewer
│   │   │   ├── Registrations/      # 5-step onboarding wizard
│   │   │   │   ├── index.jsx       # Wizard container + step navigation
│   │   │   │   ├── StepHealthProfile.jsx  # Age, gender, allergies
│   │   │   │   ├── StepComorbidities.jsx  # Disease checkboxes (diabetes, etc.)
│   │   │   │   ├── StepInsurance.jsx     # Insurance provider + plan details
│   │   │   │   ├── StepFinancials.jsx    # Income, employment status, FOIR
│   │   │   │   └── StepEmergencyContact.jsx  # Phone + alternate contact
│   │   │   ├── LoanHistory/        # Loan application tracking
│   │   │   │   └── index.jsx       # Tabular view with status badges
│   │   │   ├── Login/              # Auth entry point
│   │   │   │   └── index.jsx       # Phone/email login form
│   │   │   └── PFLDashboard/       # (Officer dashboard - see officer/ build)
│   │   │       └── index.jsx       # Loan applications review interface
│   │   ├── components/             # Shared UI components
│   │   │   ├── AppShell.jsx        # Layout wrapper (sidebar + content)
│   │   │   ├── Sidebar.jsx         # Navigation menu + user profile
│   │   │   ├── Topbar.jsx          # Header with title + actions
│   │   │   ├── EmergencyBanner.jsx # Red alert for emergency cases
│   │   │   ├── Modal.jsx           # Generic modal dialog
│   │   │   ├── StepIndicator.jsx   # Registration wizard step tracker
│   │   │   └── ui.jsx              # Reusable UI primitives (Button, Input, etc.)
│   │   ├── hooks/                  # Custom React hooks
│   │   │   ├── useChat.js          # Chat state + message management
│   │   │   ├── useLoan.js          # Loan application state + tracking
│   │   │   ├── useDocuments.js     # Document upload + extraction state
│   │   │   ├── useGeolocation.js   # GPS coordinates (hospital distance)
│   │   │   └── useProfile.js       # User registration + profile state
│   │   ├── store/                  # Zustand state management
│   │   │   ├── chatStore.js        # Chat messages, responses, hospital data
│   │   │   ├── userStore.js        # User profile, health, financials, documents
│   │   │   └── uiStore.js          # UI state, toast notifications, modals
│   │   ├── api/                    # Axios API client helpers
│   │   │   ├── client.js           # Axios instance with auth headers
│   │   │   ├── chat.js             # POST /api/chat endpoint
│   │   │   ├── auth.js             # Auth helpers (getUserId, isRegistered)
│   │   │   ├── session.js          # JWT + localStorage management
│   │   │   ├── registration.js     # POST/GET registration profile
│   │   │   ├── loan.js             # POST /api/loan (PFL application)
│   │   │   ├── documents.js        # POST /api/extract-document
│   │   │   ├── financials.js       # Financials API helpers
│   │   │   └── pfl.js              # PFL officer status checks
│   │   ├── utils/                  # Helper utilities
│   │   │   ├── helpers.js          # Date, string, array utilities
│   │   │   ├── formatCurrency.js   # ₹ formatting + locale
│   │   │   └── staticData.js       # Constants, enums, lookup tables
│   │   ├── styles/                 # Global CSS / Tailwind overrides
│   │   └── index.html              # HTML template (patient app)
│   ├── officer/                    # PFL officer dashboard (separate Vite build)
│   │   ├── main.jsx                # Officer app entry point
│   │   └── index.html              # Officer app HTML template
│   ├── dist/                       # Built patient app (npm run build)
│   ├── dist-pfl/                   # Built officer app (npm run build:pfl)
│   ├── vite.config.js              # Patient app Vite config (port 5173)
│   ├── vite.pfl.config.js          # Officer app Vite config (port 5174)
│   ├── eslint.config.js            # ESLint rules
│   ├── vercel.json                 # Vercel deployment config (SPA routing)
│   ├── package.json                # Dependencies + build scripts
│   ├── package-lock.json           # Lock file
│   └── README.md                   # Frontend-specific docs
│
├── railway.toml                    # Railway.app deployment config
└── requirements.txt                # Python backend dependencies

Testing & Debugging

Test Intent Extraction

cd backend
python test_intent.py

Directly test the Gemini intent extraction node with sample symptoms.

View API Docs

FastAPI Swagger UI: http://localhost:8000/docs
FastAPI ReDoc: http://localhost:8000/redoc

Debug Mode

Add to .env:

DEBUG=true
LOG_LEVEL=DEBUG

Common Issues

"Supabase connection failed"

Check SUPABASE_URL and SUPABASE_KEY in .env
Verify your Supabase project is active

"Gemini API key invalid"

Ensure GEMINI_API_KEY is set and has access to Gemini 2.5 Flash model
Check API quotas in Google Cloud console

"Hospital data not loading"

Ensure CSVs exist in backend/data/
Check Supabase tables are seeded: users, hospitals, procedures, cities

Frontend won't connect to backend

Verify backend is running on http://localhost:8000
Check VITE_API_BASE_URL in .env.local
Browser console should show no CORS errors

Database Schema

Key Supabase tables:

Table	Columns	Purpose
`users`	id, phone, email, profile_json, created_at	User accounts + profiles
`hospitals`	id, name, city, rating, nabh_accred, procedures	Hospital directory
`procedures`	id, icd10_code, name, base_cost, avg_cost	Procedure catalog
`cities`	id, name, latitude, longitude	City location data
`loans`	id, user_id, amount, status, cibil_score	Loan applications
`documents`	id, user_id, type, url, extracted_json	Uploaded PDFs + extraction
`chats`	id, user_id, intent, response_json, timestamp	Chat history

Roadmap

_{Built by Team NightLamps}

Name		Name	Last commit message	Last commit date
Latest commit History 64 Commits
backend		backend
medpathai		medpathai
.env.example		.env.example
.gitignore		.gitignore
README.md		README.md
railway.toml		railway.toml
requirements.txt		requirements.txt

Folders and files

Latest commit

History

Repository files navigation

What is MedPath AI?

Features

Coming Soon

Tech Stack

Getting Started

Prerequisites

1. Clone the repo

2. Backend — Python environment

3. Backend — environment variables

4. Frontend — environment variables

5. Run the backend

6. Frontend

7. Verify everything is running

Build & Deployment

Build for production

Deploy to Vercel (frontend)

Deploy to Railway (backend)

API Endpoints

Chat & Recommendations

Registration & Profile

Loan & Finance

Document Processing

Security & Auth

Security Features

How It Works

ML Scoring Pipeline

Project Structure

Testing & Debugging

Test Intent Extraction

View API Docs

Debug Mode

Common Issues

Database Schema

Roadmap

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages