armenian-labor-law-rag
haykminasyan-devAI/armenian-labor-law-rag
Retrieval-Augmented Generation (RAG) system for Armenian Labor Law Q&A. Implements BM25, Dense (Armenian embeddings), and Hybrid retrieval strategies with evaluation pipeline. Features NVIDIA Llama 3.1-70B for generation, achieving 100% Armenian responses with zero hallucination rate.
Summary
A Retrieval-Augmented Generation (RAG) system designed to answer questions about Armenian Labor Law in Armenian. It implements hybrid retrieval (BM25 + dense embeddings), uses specialized Armenian text embeddings, and interfaces with large language models via the NVIDIA API to generate accurate, citation-grounded answers. The project includes a Streamlit web interface, an evaluation pipeline, and demonstrates significant improvements in reducing hallucinations and improving citation accuracy.