AI-powered healthcare cost estimator — enter your symptoms and insurance plan, and ClearCare finds nearby providers, checks your network status, and calculates your real out-of-pocket cost before you ever step foot in a clinic.

1. Describes symptoms — via text, voice recording, or insurance card upload
2. Finds nearby providers — searches the CMS NPI Registry using your zip code
3. Checks network status — determines in-network vs out-of-network for your plan
4. Calculates out-of-pocket cost — applies your actual deductible, copay, and coinsurance
5. Generates an AI summary — plain-English explanation with a specific recommended next step
6. Speaks the summary aloud — ElevenLabs TTS for hands-free use

clearcare/
├── backend/
│   ├── main.py                  # FastAPI app, CORS, startup
│   ├── config.py                # Central config — all env vars loaded here
│   ├── railway.json             # Railway deployment config
│   ├── requirements.txt
│   ├── agent/
│   │   ├── graph.py             # LangGraph pipeline (main agent)
│   │   ├── critique.py          # Self-critique and rewrite loop
│   │   ├── tools.py             # LangChain tools (cost, network, search)
│   │   ├── prompts.py           # LLM prompt templates
│   │   ├── analytics.py         # Supabase logging with retry logic
│   │   └── memory.py            # Session/context memory
│   └── routes/
│       ├── estimate.py          # POST /api/estimate/ — main agent route
│       ├── voice.py             # POST /api/voice/transcribe, /speak
│       └── image.py             # POST /api/image/extract — insurance card OCR
│
└── frontend/
    └── app/
        ├── page.tsx             # Main page — layout and state
        ├── globals.css          # All styles
        └── components/
            ├── InputPanel.tsx       # Symptom + insurance text input
            ├── VoiceInput.tsx       # Microphone recording
            ├── InsuranceUpload.tsx  # Insurance card image upload + OCR
            ├── ResultsPanel.tsx     # Cost breakdown and confidence score
            ├── InsuranceSavings.tsx # Insurance savings card
            ├── AIAnalysis.tsx       # AI summary and recommended next step
            ├── HospitalCards.tsx    # Nearby provider cards
            ├── HospitalMap.tsx      # Google Maps provider map
            ├── Header.tsx
            └── Footer.tsx

The backend runs a LangGraph multi-step agent for each request:

map_symptoms → assess_severity → find_hospitals → check_network
     → estimate_cost → find_alternatives → generate_answer → critique

• Python 3.11+
• Node.js 18+
• A Supabase project
• API keys (see below)

cd backend
python -m venv venv
venv\Scripts\activate        # Windows
pip install -r requirements.txt

Copy .env.example to .env and fill in the values:

cp .env.example .env

Start the server:

uvicorn main:app --reload --port 8000

cd frontend
npm install

Create frontend/.env.local:

NEXT_PUBLIC_API_URL=http://localhost:8000
NEXT_PUBLIC_GOOGLE_MAPS_API_KEY=your_key_here

Start the dev server:

npm run dev

Open http://localhost:3000.

Note: Use a standard sk- OpenAI key in production. Project-scoped sk-proj- keys can have IP restrictions that block cloud hosting providers like Railway.

1. Connect your GitHub repo to Railway
2. Set root directory to backend/
3. Add all backend environment variables in Railway → Variables
4. Railway uses railway.json to build and start with uvicorn

1. Connect your GitHub repo to Vercel
2. Set root directory to frontend/
3. Add NEXT_PUBLIC_API_URL pointing to your Railway backend URL (no trailing slash)
4. Add NEXT_PUBLIC_GOOGLE_MAPS_API_KEY

Every query is logged to Supabase (clearcare_queries table) with:

• Session ID, symptoms, zip code, insurance plan
• Hospitals found, confidence score (before and after self-critique)
• Urgency level, whether defaults were used
• Timestamp

This data is modelled and visualised in Lightdash.

The lightdash-supabase-quickstart/ folder contains the full analytics stack as code — models, charts, and a dashboard — all connected to the Supabase database.

clearcare_queries — one row per user query

sessions — one row per user session (linked to queries)

cd lightdash-supabase-quickstart

Set your Supabase credentials in the warehouse config, then deploy:

lightdash deploy --project <your-project-uuid>

Or push charts and dashboards with:

lightdash upload --force