From Naive RAG to Deep Agentic Retrieval: An Evolving Context Engineering Pipeline for Regulatory Compliance

A new study published on arXiv (2607.24791) discusses a retrieval system developed by Ontario Power Generation (OPG) designed to ensure compliance with regulations and facilitate rate case analysis based on Ontario Energy Board (OEB) standards. The authors describe several stages, such as naive retrieval augmented generation (RAG), hybrid methods with re-ranking, and a sophisticated multi-agent system that includes tool synthesis and detailed planning. They address the challenges that led to these advancements. The system, called Progressive Evidence Acquisition with Cost-Aware Escalation (PEA-CAE), starts with affordable, precise retrieval, moving to comprehensive document reviews only when necessary. The research points out the shortcomings of basic techniques as the size of the data and complexity of queries increase, highlighting RAG as the optimal method for handling enterprise document collections with LLMs.