Towards Data Science is a community-powered publication that showcases work in data science, machine learning and artificial intelligence. Every day newcomers, seasoned researchers and industry practitioners publish tutorials, research notes and real-world case studies that help the field move forward.

Towards Data Science

Proxy-Pointer RAG is a novel retrieval pipeline that achieves the structural accuracy of Vectorless RAG (PageIndex) at the cost and scale of standard Vector RAG. It uses five zero-cost engineering techniques: building a skeleton tree via regex (no LLM), injecting breadcrumb ancestry paths into embeddings for structural awareness, structure-guided chunking that respects section boundaries, metadata pointers that retrieve full document sections instead of raw chunks, and noise filtering to remove distracting sections like table of contents. On a 10-query benchmark against a 131-page World Bank report, Proxy-Pointer matched or beat PageIndex on 8 out of 10 queries while requiring zero LLM calls at indexing time and no tree-navigation step at retrieval — matching standard Vector RAG in cost and scalability.

Proxy-Pointer RAG: Achieving Vectorless Accuracy at Vector RAG Scale and Cost

Comparison of Vectorless vs Flat Vector RAG

Engineering a Better Retriever — Proxy-Pointer RAG