CS210 Autograder Dashboard

Instructor-facing dashboard to grade PRs against team PRDs.

Made with Cursor, based on my PRD

What This Includes

• Team selector for CS210 project repos
• PRD upload (file or Google Drive link)
• Overall progress summary
• P0/P1 recommendations
• PR summaries timeline (score + summary per PR)
• Force refresh and nightly backfill actions
• "Talk to Autograder (Experimental)" panel

Quick Start (Recommended: Docker)

1) Clone and configure env

git clone <your-repo-url>
cd 210_autograder
cp backend/.env.example backend/.env

Edit backend/.env and set at minimum:

• GITHUB_TOKEN=<token-with-access-to-cs210-private-repos>

Optional for better summaries:

• OPENROUTER_API_KEY=<your-key> and LLM_PROVIDER=openrouter

2) Run

docker compose up --build

Open:

• http://127.0.0.1:8000

Docker stores SQLite data in a persistent named volume (autograder_data).

3) Stop

docker compose down

Local Python Run (No Docker)

cd backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
uvicorn app.main:app --reload

Open:

• http://127.0.0.1:8000

Required Access and Tokens

• GITHUB_TOKEN is required for most CS210 team repos (private org repos).
• If force refresh fails with repo access errors, verify token permissions and org authorization.

Recommended token permissions:

• Private repo read access
• Pull requests read
• Contents read
• Metadata read

LLM Options (Free-Friendly)

Choose one:

1. OpenRouter free model (easiest)

• LLM_PROVIDER=openrouter
• OPENROUTER_API_KEY=<key>
• OPENROUTER_MODEL=meta-llama/llama-3.1-8b-instruct:free (default)

2. Local Ollama

• Install Ollama app
• Run ollama pull llama3.1
• Keep Ollama running on http://localhost:11434
• Set LLM_PROVIDER=ollama

If no LLM is reachable, app falls back to heuristic summaries and shows warning text.

Environment Variables

See backend/.env.example.

Important values:

• GITHUB_TOKEN
• MAX_PRS_PER_TEAM (default 40)
• AUTO_IMPORT_CS210_TEAMS (default true)
• LLM_PROVIDER (auto, openrouter, ollama, heuristic)
• OPENROUTER_API_KEY
• OPENROUTER_MODEL
• OLLAMA_URL
• OLLAMA_MODEL

Admin Onboarding (Suggested)

For each new admin:

1. Pull latest code
2. Copy env and add their own token
3. Start with Docker (docker compose up --build)
4. Open dashboard and click Force Refresh on a team
5. Confirm summaries populate

Health Check

• GET /healthz returns {"status":"ok"}

Notes for Contributors

• Backend: FastAPI + SQLite
• Main entrypoint: backend/app/main.py
• Core grading logic: backend/app/services.py
• GitHub/LLM integrations: backend/app/integrations.py