AI Service (ai_med_extract)
===========================

Quick dev notes for the migrated `ai_med_extract` service.

Run a lightweight smoke test (does not initialize models):

```powershell
python run_smoke_test.py
```

Create the Flask app without loading models:

```python
from ai_med_extract import create_app
app = create_app()

# The app is ready for tests; call initialize_agents(app, preload_small_models=False)
# to wire routes with lazy/fallback agents that avoid downloads.
```

Start dev server and initialize agents (will attempt to load small models):

```powershell
# python -c "from ai_med_extract.app import create_app, initialize_agents; app=create_app(); initialize_agents(app, preload_small_models=True); app.run(host='0.0.0.0', port=7860)"
```

Environment flags of interest:
- `HF_SPACES=true` — signals HF Spaces environment
- `PRELOAD_GGUF=true` — will enable preloading GGUF in initialization (disabled by default)
- `HF_HOME`, `TORCH_HOME`, `WHISPER_CACHE` — cache locations

Documentation and deployment (k8s, prometheus) should be added to this file when ready.