Proxy-Pointer RAG: Structure Meets Scale at 100% Accuracy with Smarter Retrieval

Proxy-Pointer RAG is a retrieval architecture that dramatically improves accuracy by preserving a document's inherent structure within the vector index. Instead of feeding language models fragmented text, this system uses retrieved chunks as "pointers" to load the complete, unbroken document section, ensuring full context. A refined two-stage process first performs a broad semantic search and then uses an LLM to re-rank results based on structural relevance, prioritizing the most logical sections over simple keyword matches. When stress-tested on 66 complex financial questions across dense 10-K reports, the open-source pipeline achieved a perfect 100% accuracy score.