Deployment Blueprint

GitHub + Vercel + Hugging Face architecture

The main web product is now the Next.js app in apps/web. Vercel deploys that app, and the browser only communicates with internal route handlers. Hugging Face access stays on the server side.

Release posture

Minimal architecture, stronger production story.

The platform is optimized for portfolio review and committee demonstrations while preserving a secure boundary between the browser and the inference backend.

Request flow
Browser
  -> Next.js page/app router
  -> POST /api/jobs
  -> Hugging Face Space (server-side only) or mock scorer
  -> Normalized JSON + downloadable CSV
Key environment variables
INFERENCE_BACKEND=mock | hf_space
HF_SPACE_ID=Pattarabordee/pea-ne1-meter-detection
HF_API_NAME=/predict
HF_AUTH_TOKEN=
HF_USE_AUTH=false
ENABLE_MOCK_FALLBACK=true
Why this is Vercel-ready
  • No dependency on a separate Go service for the main demo path
  • No reliance on in-memory server jobs between requests
  • Next.js route handlers are the only public API surface needed for the demo
Why this is public-safe
  • Uses synthetic sample input only
  • Does not expose HF tokens to the browser
  • Includes mock mode for local demos and fallback resilience