Corrective RAG System with Self-Healing Retrieval & Hallucination Detection

An advanced RAG system with a self-correction layer that evaluates the quality of retrieved documents before generating answers — minimizing hallucinations and improving accuracy. Built around Ancient Egyptian history as a knowledge domain, demonstrating reliable AI over specialized, niche content.

Key highlights:

- LLM-based document evaluator that classifies retrieved chunks as Correct, Ambiguous, or Incorrect

- Automatic query rewriting triggered when retrieval quality is insufficient

- Semantic chunking with sliding window fallback and spaCy-based knowledge refinement

- Tavily Search API fallback when local knowledge base is insufficient

- Deployed live on Hugging Face Spaces

Source code: https://github.com/SaraaE...