README.md

---
title: Arabic Complaint Classification API
emoji: 🔍
colorFrom: blue
colorTo: green
sdk: docker
app_port: 7860
---

Production-grade Arabic complaint classification: three MARBERT-v2 classifiers (sentiment, topic, intent) plus a deterministic routing engine and optional LLM explanation layer.

**Source code:** [github.com/ysfxjo55/arabic-complaint-classification-api](https://github.com/ysfxjo55/arabic-complaint-classification-api)

## API

| Endpoint | Method | Description |
|----------|--------|-------------|
| `/predict` | POST | Classify a complaint (sentiment, topic, intent, action) |
| `/explain-classification` | POST | Same as `/predict` plus optional LLM explanation |
| `/health` | GET | Liveness check |
| `/ready` | GET | Readiness check (models loaded) |
| `/docs` | GET | OpenAPI Swagger UI |

### Example

```bash
curl -X POST https://Ysfxjo-arabic-complaint-classification.hf.space/predict \
  -H "Content-Type: application/json" \
  -d '{"text": "التطبيق يعلق عند الدفع ومحدش يرد عليّ"}'
```

Set `HF_TOKEN` in Space secrets so the container can download the three MARBERT models from Hugging Face Hub on startup.