---
title: SEMA Web Demo
sdk: docker
app_port: 7860
pinned: false
---

# SEMA Web Demo

This folder contains a Hugging Face Docker Space-ready web demo for the current SEMA pipeline.

## What it does

- Accepts a user-uploaded archery video.
- Saves it only to a temporary file inside the container.
- Runs the existing SEMA analysis flow:
  - `Tokenize_SearchKeyword(...)`
  - `get_response(...)`
  - `answer_archery_question(...)`
- Deletes the uploaded file after analysis finishes.
- Keeps the evaluation context only in process memory for follow-up Q&A.

## Runtime model

- No persistent user data.
- No video archive.
- No login or settings storage.
- Single-container deployment.
- Best fit: small-scale research demo.

## Files

- `backend/app.py`: FastAPI backend, job queue, temp upload handling, follow-up chat.
- `static/`: upload/result/chat frontend.
- `Dockerfile`: container build for Hugging Face Docker Space.
- `requirements.txt`: web-only Python dependencies.

## Environment variables

- `ALI_API_KEY`: required, used by DashScope-compatible OpenAI endpoints.
- `SEMA_MAX_UPLOAD_MB`: optional, defaults to `80`.
- `SEMA_JOB_TTL_SECONDS`: optional, defaults to `7200`.
- `SEMA_MAX_WORKERS`: optional, defaults to `1`.
- `SEMA_SUBPIPELINE`: optional, defaults to `4`.

## Deploy to a Hugging Face Docker Space

1. Create a new Space and choose `Docker`.
2. Use the current repository as the source, or copy the repo contents into the Space repository.
3. Place the contents of this `webapp/` folder at the Space repo root if you want Hugging Face to use the included `README.md` and `Dockerfile` directly.
4. In Space settings, add `ALI_API_KEY` as a secret.
5. Push and wait for the container build to finish.

## Local run

From the repository root:

```bash
docker build -f webapp/Dockerfile -t sema-web .
docker run --rm -p 7860:7860 -e ALI_API_KEY=your_key sema-web
```

Then open `http://localhost:7860`.