RAGReader lets you ask questions about your own documents (PDF, TXT, Markdown, a web page URL, or pasted text) and see how well different Retrieval-Augmented Generation (RAG) setups answer them.
A normal chat answers from dense vector retrieval. A Deep Analysis then runs the same question through every combination of retrieval method and language model: Dense retrieval (embeddings), Sparse retrieval (BM25) and Hybrid retrieval (dense + sparse candidates reranked), across models such as GPT-4o mini, Gemini 3 Flash and Claude Haiku 4.5. Results stream in live with a step-by-step analysis flow.
Each answer is scored against ground truth that you choose: pick the relevant passages by hand, or let candidate pooling with Reciprocal Rank Fusion build it automatically. Retrieval is measured with Precision@K, Recall@K and F1@K; answers are measured with Ragas faithfulness, response relevance and factual correctness.
Optional RAG modules (HyDE, RAG Fusion, Step-Back Prompting, CRAG, RAPTOR, FLARE, Self-Route, Adaptive RAG and more) can be switched on to compare techniques, and a built-in course teaches RAG fundamentals, retrieval, evaluation and module composition.
RAGReader is MIT-licensed and can be self-hosted with Docker (Django, Celery, Redis, React).
No comments or reviews, maybe you want to be first?