MathLens

Project structure

MathLens/
├── frontend/                        # Next.js web app
│   └── src/
│       ├── app/                     # Pages and layout
│       ├── components/              # Upload, camera, LaTeX preview
│       └── lib/                     # API client (calls backend)
└── backend/
    ├── app/
    │   ├── main.py                  # FastAPI routes
    │   ├── errors.py                # Gemini / LangChain error mapping
    │   └── services/
    │       └── latex_extractor.py   # LangChain + Gemini vision integration
    ├── requirements.txt             # langchain-core, langchain-google-genai
    ├── Dockerfile                   # Lambda container image
    ├── lambda_handler.py            # Mangum ASGI adapter
    └── template.yaml                # AWS SAM deployment template

Deployment

Frontend

• Where: Vercel → mathlens.npsolver.io
• How: Import repo in Vercel with root directory frontend
• DNS: Cloudflare CNAME mathlens → cname.vercel-dns.com
• Env: API_SERVICE_URL=https://api.mathlens.npsolver.io

Backend (FastAPI)

• Where: AWS Lambda (container image) behind API Gateway HTTP API → api.mathlens.npsolver.io
• How: AWS SAM (sam build && sam deploy) from backend/
• DNS: ACM certificate + API Gateway custom domain + Cloudflare CNAME api.mathlens → API Gateway target domain

LangChain + Gemini

• Where: LangChain runs inside the backend Lambda (backend/app/services/latex_extractor.py); it is not deployed as a separate service
• How: langchain-google-genai calls Google Gemini (vision) via ChatGoogleGenerativeAI and HumanMessage (text prompt + base64 image)
• Model: Google Gemini (GEMINI_MODEL, default gemini-2.5-flash) via Google AI Studio
• Credentials: GOOGLE_API_KEY set as a Lambda environment variable in backend/template.yaml