medo832/islamic-egypt-history

# Islamic Egypt History — Arabic RAG System An Arabic Retrieval-Augmented Generation (RAG) system that answers questions about the history of Islamic Egypt, grounded in a curated Arabic corpus. Large language models often hallucinate when asked about specialized Arabic historical topics — dates get invented, names get confused, and sources get fabricated. This project addresses that by retrieving relevant passages from a verified Arabic Wikipedia corpus on the Islamic conquest of Egypt and instructing the model to answer **only** from those passages, admitting honestly when the answer is not in the corpus. **🔴 Live demo:** medo9090-islamic-egypt-hist… ## Architecture ``` data.txt (Arabic corpus) │ ▼ Chunking ── whitespace word split, accumulated into ~400-character chunks │ ▼ Embeddings ── sentence-transformers/paraphrase-multilingual-mpnet-base-v2 │ ▼ FAISS IndexFlatL2 ── exact L2 vector index, built in memory at startup │ ▼ Retrieval ── query embedded with the same model, top-5 nearest chunks │ ▼ Prompt construction ── Arabic instruction: answer ONLY from the retrieved │ passages; if no clear answer exists, say so honestly ▼ Generation ── Groq API, llama-3.3-70b-versatile, temperature 0.3 │ ▼ Gradio UI ── question textbox in, answer textbox out ``` All steps run inside `app.py` (the deployed Hugging Face Space). `main.py` is the earlier development script used to prototype the pipeline (smaller chunks, top-3 retrieval) and runs the same core stages from the command line. ## Tech Stack | Component | Choice | |---|---| | Embeddings | `paraphrase-multilingual-mpnet-base-v2` (Sentence Transformers) | | Vector store | FAISS (`IndexFlatL2`) | | LLM | Llama 3.3 70B via Groq API | | UI | Gradio | | Language | Python | ## Installation ```bash git clone github.com cd islamic-egypt-history pip install -r requirements.txt ``` Create a `.env` file in the project root with your Groq API ke …