| --- |
| title: Document Intelligence Demo |
| emoji: "📄" |
| colorFrom: gray |
| colorTo: blue |
| sdk: docker |
| app_port: 7860 |
| pinned: false |
| license: mit |
| short_description: Hybrid BM25 + MiniLM retrieval with cited passages. No LLM. |
| --- |
| |
| # Document Intelligence demo API |
|
|
| This Space runs the API behind the public demo of |
| [document-intelligence-ai](https://github.com/cbratkovics/document-intelligence-ai). |
| It is deployed automatically from that repository's `main` branch; do not |
| edit files here. |
|
|
| ## What it does |
|
|
| - Hybrid retrieval over a small fictional corpus: BM25 fused with |
| `sentence-transformers/all-MiniLM-L6-v2` embeddings (ONNX Runtime via |
| fastembed) using reciprocal rank fusion. A heuristic term-overlap reranker |
| is available on request. |
| - Every hit reports its BM25 rank, dense rank, fused rank and score |
| separately, so a client can show how the two retrievers disagree. |
| - Retrieval only. `GENERATION_PROVIDER=none`: no language model is called |
| and no provider SDK is installed in the image. |
|
|
| ## Configuration |
|
|
| Runtime settings come from `demo.env` (committed, no secrets). The only |
| secret is `API_KEY`, set in the Space settings; every `/api/v1` route |
| requires it in the `X-API-Key` header. `/health` is open and reports the |
| effective modes and document count. |
|
|
| Limits: uploads of `.txt`, `.md`, `.rst`, `.pdf` up to 4 MB; at most 24 |
| documents in the corpus with oldest-first eviction of uploads (the four |
| seeded samples are never evicted); per-client and global rate limits; |
| storage is ephemeral and resets whenever the Space restarts. |
|
|
| ## Endpoints used by the demo frontend |
|
|
| `GET /health`, `POST /api/v1/search`, `POST /api/v1/documents/upload`, |
| `GET /api/v1/documents`, `GET /api/v1/documents/{id}/chunks`. |
| Full OpenAPI schema at `/docs`. |
|
|
