docs/hugging-face.md

Hugging Face Spaces Deployment

PolicyTrace should be deployed as a Docker Space because it is a FastAPI plus React application, not a pure Gradio or Streamlit app.

Deployment Shape

The root Dockerfile does this:

1. Builds the React UI with Vite.
2. Installs the Python backend dependencies.
3. Downloads the small spaCy English model used by Presidio.
4. Copies ui/dist into the image.
5. Starts FastAPI on port 7860.
6. Lets FastAPI serve both /api/* and the React app.

Space Settings

Create a new Hugging Face Space:

• SDK: Docker
• Port: 7860
• Visibility: public or private, depending on your demo plan

Add this secret in the Space settings:

GROQ_API_KEY=your_groq_key

Optional secrets or variables:

GROQ_MODEL=meta-llama/llama-4-scout-17b-16e-instruct
GROQ_CLASSIFIER_MODEL=llama-3.1-8b-instant

Public Demo Safety

For a public Space, use only the synthetic PDFs in:

sample_data/policytrace_demo_pack/

Do not upload real customer documents to a public demo unless you have explicit permission and strong retention controls.

Storage Notes

Hugging Face Spaces have ephemeral storage by default. This means generated sessions may disappear when the Space restarts.

For a public portfolio demo, ephemeral storage is usually fine. For a persistent review workflow, enable persistent storage or move sessions to an external object store/database.

Local Docker Test

Before pushing to a Space:

docker build -t policytrace .
docker run --rm -p 7860:7860 --env-file .env policytrace

Then open:

http://localhost:7860

Linking From This Repo

After the Space is live, add the Space URL to the main README.md demo section.