ogonkem/naija-civic-guard

Nigerian Constitution RAG (Retrival-Augumented Generation) This is the final, comprehensive version of your README.md. It now includes a "Performance Benchmarks" section that highlights the actual metrics from your eval_report.md, making the project look highly data-driven and professional. ------------------------------ ## 🇳🇬 Naija Civic Guard: Nigerian Constitution RAG Empowering citizens with AI-driven, verifiable insights into the Nigerian Constitution. Naija Civic Guard is a high-precision Retrieval-Augmented Generation (RAG) system. Unlike generic LLMs, this tool is strictly grounded in the Official Gazette of the Nigerian Constitution. It utilizes a Hybrid Retrieval strategy and a rigorous evaluation pipeline to provide reliable legal answers with direct source citations. ------------------------------ ## 📊 Performance Benchmarks We don't guess—we measure. Based on our latest automated evaluation against the evaluation_set.jsonl, the system achieves the following metrics: | Metric | Value | Description | |---|---|---| | Hit Rate (Top-K) | 60% | The correct legal section is retrieved in the top results 6 out of 10 times. | | Avg Keyword Coverage | 84% | Our responses maintain high legal accuracy by including essential terminology. | | Mean Reciprocal Rank | 0.45 | Measures how highly the system ranks the specific target Section. | Continuous testing ensures that updates to the retrieval weights or chunking strategies improve these scores over time. ------------------------------ ## 🏗️ Architecture & Pipeline## 1. Hybrid Ingestion (ingest.py) Our ingestion process is a "legal-aware" pipeline designed for maximum precision: * Metadata Tagging: Uses Regex to extract and tag chunks with specific Section and Article numbers for granular citations. * Ensemble Retrieval: Combines Vector Search (ChromaDB) for semantic intent and Keyword Search (BM25) for specific legal terms. This ensures that a query about "Fundamental Rights" finds Section 33 even if the exact wording varies. ## 2. RAG Service (services.py) The RagServi …