nahomWM/Ethiopian-Legal-Answer-RAG

A high-performance Retrieval-Augmented Generation (RAG) system designed specifically for Ethiopian legal document processing,from crawling legislative sites to generating context-aware legal answers. # Ethiopian Legal Answer RAG 🇪🇹 A high-performance **Retrieval-Augmented Generation (RAG)** system designed specifically for Ethiopian legal document processing. This project handles the entire pipeline—from crawling legislative sites to generating factual, context-aware legal answers. ### 💨 Quick Demo **Query**: *"What does the constitution say about sovereignty?"* **Response**: *"According to Article 8 of the Ethiopian Constitution, all sovereign power resides in the Nations, Nationalities and Peoples of Ethiopia. This sovereignty is expressed through their elected representatives and direct democratic participation."* ## 🌟 Key Features - **Amharic NLP Suite**: Custom Geez script normalization, deep text cleaning, and recursive Amharic-optimized chunking. - **Multi-Lingual Embeddings**: Uses **BGE-M3**, the state-of-the-art model for Amharic and English cross-lingual retrieval. - **Hybrid LLM Support**: - **Local (Zero Cost)**: Native integration with **Ollama** (Llama-3, Mistral). - **Cloud (Free)**: Integration with **Hugging Face Inference API** for cloud-based RAG. - **Infrastructure**: Pre-configured **Kafka** (data streaming) and **Qdrant** (vector search) environment. --- ## 🏗️ Project Architecture ```text services/ scraper/ - Resilient crawlers for HOPR and legal sites. processor/ - NLP processing, BGE embeddings, and vector upsert logic. engine/ - LangChain-powered RAG chains with expert legal templates. libs/ shared/ - Centralized Pydantic v2 settings and Geez NLP utilities. infra/ - Containerized persistence layer (Qdrant & Kafka). main.py - Unified CLI entry point for all operations. ``` --- ## � Setup & Installation ### 1. Prerequisite Environments ```bash python3 -m venv venv source venv/bin/activate pip install -r requirements.txt ``` ### 2. Launch Infrastructure Start the vector store and message bus: ```bash cd infra && docker-compose up -d ``` ### 3. Configure Your "Brain" (LLM) #### **Option A: Completely Local (R …