Mediscribe SaaS

AI-powered consultation assistant for doctors in private practice.

This SaaS application turns raw consultation notes — typed or dictated — into professional medical record summaries, actionable next steps, and patient-friendly emails, with one-click translation into 19 languages.

---

Features

• Professional summaries — generates structured medical record summaries from consultation notes
• Action items — extracts clear next steps and follow-up actions for the doctor
• Patient emails — drafts patient-friendly email communications automatically
• Audio dictation — record consultation notes by voice; transcribed instantly via OpenAI Whisper
• Multilingual letters — translate patient emails into 19 languages with one click
• Real-time streaming — AI output streams token-by-token as it's generated
• Subscription gating — Clerk-powered authentication and plan protection

---

Tech Stack

Layer	Technology
Frontend	Next.js (Pages Router), TypeScript, Tailwind CSS
Backend	FastAPI (Python), Uvicorn
AI	OpenAI GPT-4o-mini (summaries, translation), Whisper (transcription)
Auth	Clerk (authentication + subscription management)
Deployment — Vercel	Vercel (frontend) + Vercel Serverless Functions (API)
Deployment — AWS	Docker, AWS ECR, AWS App Runner

---

Project Structure

saas/
├── api/
│   ├── index.py        # Vercel: FastAPI app (summary, transcribe, translate)
│   └── server.py       # AWS: FastAPI app + static file serving
├── pages/
│   ├── _app.tsx        # ClerkProvider wrapper
│   ├── _document.tsx   # HTML shell
│   ├── index.tsx       # Landing page
│   └── product.tsx     # Main consultation form (protected)
├── styles/
│   └── globals.css
├── public/
├── Dockerfile          # Multi-stage build (Next.js + Python)
├── requirements.txt    # Python dependencies
├── package.json        # Node dependencies
├── next.config.ts      # Static export config (for AWS)
└── vercel.json         # API route rewrites (for Vercel)

---

Getting Started

Prerequisites

• Node.js 18+
• Python 3.12+
• A Clerk account
• An OpenAI API key

1. Clone the repo

git clone https://github.com/mrithwik/saas.git
cd saas

2. Install dependencies

# Frontend
npm install

# Backend
pip install -r requirements.txt

3. Set up environment variables

Create a .env.local file in the project root:

NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY=pk_test_...
CLERK_SECRET_KEY=sk_test_...
CLERK_JWKS_URL=https://your-clerk-app.clerk.accounts.dev/.well-known/jwks.json
OPENAI_API_KEY=sk-...

Never commit .env.local or .env to version control.

4. Run locally (development)

# Frontend (Next.js dev server)
npm run dev

# Backend (FastAPI — in a separate terminal)
uvicorn api.index:app --reload --port 8000

Open http://localhost:3000 for the frontend.

---

Deployment

Vercel

1. Push to GitHub — Vercel auto-deploys on every commit to main
2. Add environment variables in Vercel project settings
3. The vercel.json rewrites route /api/* calls to the FastAPI serverless function

AWS (Docker + App Runner)

Full step-by-step instructions are in [week1/day5.md](week1/day5.md).

Quick reference — rebuild and redeploy after code changes:

# Load env vars
export $(cat .env | grep -v '^#' | xargs)

# Build for Linux/AMD64 (required for AWS — critical for Apple Silicon Macs)
docker build \
  --platform linux/amd64 \
  --build-arg NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY="$NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY" \
  -t consultation-app .

# Push to ECR
docker tag consultation-app:latest $AWS_ACCOUNT_ID.dkr.ecr.$DEFAULT_AWS_REGION.amazonaws.com/consultation-app:latest
docker push $AWS_ACCOUNT_ID.dkr.ecr.$DEFAULT_AWS_REGION.amazonaws.com/consultation-app:latest

Then go to AWS App Runner → select your service → Deploy.

---

API Endpoints

Method	Path	Description
POST	/api/consultation	Stream summary, next steps, and patient email
POST	/api/transcribe	Transcribe audio dictation via Whisper
POST	/api/translate	Stream translated patient letter
GET	/health	Health check (used by App Runner)

All endpoints require a valid Clerk JWT in the Authorization: Bearer <token> header.

---

Supported Languages (Translation)

French · Spanish · Mandarin Chinese · Arabic · Hindi · Portuguese · German · Japanese · Korean · Italian · Russian · Polish · Ukrainian · Farsi · Turkish · Vietnamese · Tagalog · Punjabi · Urdu

---

Environment Variables Reference

Variable	Required	Description
NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY	Yes	Clerk publishable key (baked into frontend build)
CLERK_SECRET_KEY	Yes	Clerk secret key (backend only)
CLERK_JWKS_URL	Yes	Clerk JWKS URL for JWT verification
OPENAI_API_KEY	Yes	OpenAI API key for GPT-4o-mini and Whisper
DEFAULT_AWS_REGION	AWS only	AWS region (e.g. us-east-1)
AWS_ACCOUNT_ID	AWS only	12-digit AWS account ID

---

Contributing

Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request.

---

License

This project is licensed under the MIT License — see LICENSE for details.