---
title: Arabic RAG with DSPy
emoji: 🧠
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.33.0
app_file: app.py
pinned: false
---

# 🧠 Arabic RAG System with DSPy + Gradio

This is a full Hugging Face Space project that allows:

- 🧾 Uploading Arabic PDF documents.
- 🧠 Storing and indexing chunks using ChromaDB and SentenceTransformers.
- ❓ Asking questions and generating answers using DSPy with context retrieval.
- ⚙️ Improving answer accuracy using MIPROv2 optimization based on `trainset.jsonl` and `valset.jsonl`.

---

## 🚀 How to Use

1. Go to the **"📥 تحميل وتخزين"** tab to upload your Arabic PDF file.
2. Go to the **"❓ سؤال"** tab to type a question.
3. (Optional) Go to the **"⚙️ تحسين النموذج"** tab to upload training and validation sets for improving answer accuracy.

---

## 📁 Project Structure

| File              | Description                                             |
|-------------------|---------------------------------------------------------|
| `app.py`          | Main Gradio interface and DSPy logic                    |
| `requirements.txt`| Python dependencies                                     |
| `trainset.jsonl`  | Example training data (question/answer pairs)           |
| `valset.jsonl`    | Example validation data (question/answer pairs)         |

---

## ✨ Notes

- This project supports full Arabic interaction and is optimized for educational and research purposes.
- You can use any Arabic-compatible LLM via DSPy (e.g., OpenAI, OpenChat, Mistral).

---

## 📚 Resources

- [DSPy Official Documentation](https://dspy.ai)
- [Gradio Docs](https://www.gradio.app)
- [Hugging Face Spaces Guide](https://huggingface.co/docs/hub/spaces)