Show HN: Semantic Coverage – A tool to visualize RAG blind spots using UMAP (github.com)

A new tool called "Semantic Coverage" has been launched to enhance the observability of Retrieval Augmented Generation (RAG) systems by visualizing knowledge gaps and potential hallucination areas within vector databases. Traditionally, AI engineers have operated without a clear understanding of their systems' semantic limitations, leaving them unaware of critical blind spots—queries that the knowledge base lacks context for, and how user intent may shift over time. This tool employs UMAP for dimensionality reduction and HDBSCAN for clustering, projecting both user queries and documents into a shared latent space to identify high-density user interaction areas that coincide with low document density, labeled as "Red Zones." The significance of Semantic Coverage lies in its ability to automate the detection of knowledge gaps, enabling engineers to proactively address issues that may hamper user experiences and solution accuracy. Using a backend powered by FastAPI and a frontend built on React with Plotly.js, it provides a straightforward interface for users to paste their JSON exports of queries and documents, generating a Gap Report that highlights missing topics. Additionally, it features an extensible architecture compatible with major vector databases like Pinecone and ChromaDB, making it a versatile asset for the AI/ML community focused on improving RAG systems.