A RAG-based document QA system with a Redis-powered semantic cache that caches query embeddings and generated answers using vector similarity instead of exact matching. In this a similarity router decides whether to reuse cached responses, reuse retrieved documents, or run full retrieval and generation.