Product for Teachers

Vercel-ready demo for a teacher-facing AI grading workflow.

Run Locally

npm install
cp .env.example .env.local
npm run dev

Open http://localhost:3000.

Demo Flow

1. Home page: shows the photographed student paper and the uploaded answer/scoring standard.
2. Review page: shows each question crop on the left and a simple teacher confirmation panel on the right.
3. Results page: shows the annotated paper, score summary, and export buttons.

Hardware API

The hardware partner can submit captured image metadata to:

POST /api/hardware/upload

See docs/architecture/hardware-api.md for the payload.

Backend API

This project uses Next.js Route Handlers as a lightweight backend on Vercel:

• POST /api/uploads/paper: accepts a local image file from the browser. In production, configure BLOB_READ_WRITE_TOKEN to store files in Vercel Blob. Without it, local development returns an inline preview URL.
• POST /api/grade: calls Kimi from the server using KIMI_API_KEY.
• POST /api/hardware/upload: receives image metadata from a hardware partner.
• POST /api/worker/callback: receives crop/OCR results from the Python worker.

Vercel environment variables:

KIMI_API_KEY=your_kimi_key
KIMI_MODEL=moonshot-v1-8k
BLOB_READ_WRITE_TOKEN=your_vercel_blob_token

Processing Boundary

Vercel handles the web UI, upload endpoint, job state, and callback endpoint. Expensive crop and OCR work should run in the Python worker described in workers/python/README.md.