Show HN: Vectorless, Reasoning-Based RAG

Category: ai-ml

Tags: rag, llm, document-retrieval, reasoning, tree-index

Score: 7.5/10 (Innovation: 8, Technical: 7, Documentation: 8, Utility: 7)

PageIndex is a vectorless RAG engine that replaces vector similarity search with a hierarchical tree index and LLM reasoning to retrieve relevant sections from long professional documents. It eliminates chunking and vector databases, providing traceable, context-aware retrieval that mirrors human reading and achieves state-of-the-art accuracy on FinanceBench.

Target audience: AI/ML engineers, backend developers, data scientists working with document QA and long-form retrieval

Repository: https://github.com/VectifyAI/PageIndex · Python · MIT · 39434 stars

View on Hacker News