MDKeyChunker: Single-Call LLM Enrichment with Rolling Keys and Key-Based Restructuring for High-Accuracy RAG

Researchers introduce MDKeyChunker, a three-stage pipeline that improves RAG (Retrieval-Augmented Generation) systems by using structure-aware chunking of Markdown documents, single-call LLM enrichment, and semantic key-based restructuring. The system achieves superior retrieval performance with Recall@5=1.000 using BM25 over structural chunks, significantly improving upon traditional fixed-size chunking methods.

• →MDKeyChunker performs structure-aware chunking that treats headers, code blocks, tables, and lists as atomic units rather than using fixed-size chunks.
• →The system extracts seven metadata fields in a single LLM call, eliminating the need for multiple extraction passes and improving efficiency.
• →Rolling key propagation maintains document-level context and enables semantic matching without hand-tuned scoring systems.
• →Empirical evaluation shows Config D achieves perfect Recall@5=1.000 and MRR=0.911 on a 30-query, 18-document corpus.
• →The implementation is lightweight with only four Python dependencies and supports any OpenAI-compatible endpoint.