Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces Iterative Nugget Optimization (INO), a method for automatically refining knowledge base entries ("factual nuggets") to improve their discoverability by agentic RAG systems in response to factual correction feedback. INO iteratively revises nuggets based on the agent's retrieval and answer traces when probed with the original query and paraphrases, effectively using the production RAG system as a test harness. Experiments with two real-world B2B knowledge-assistance agents demonstrate that INO significantly enhances the discoverability and utilization of factual corrections compared to baseline approaches.
Automatically refine your knowledge base entries to be more discoverable by RAG agents, using the agent itself as a test environment for iterative optimization.
Agentic retrieval-augmented generation (RAG) systems in complex B2B (business-to-business) settings may often receive free-form response feedback. Rather than generic feedback signals such as style, preference, or overall response quality, we focus on actionable factual corrections. We identify these instances and convert them into compact knowledge-base entries, which we call factual nuggets. We introduce Iterative Nugget Optimization (INO), an index-time optimization method that uses the production agentic RAG as a test harness: it creates an initial nugget, probes it with the triggering query and paraphrases, reflects over failed retrieval and answer traces, and revises the nugget until it is discoverable. We evaluate INO with two production B2B knowledge-assistance agents across multiple companies that use our system: a product support agent that answers questions over company-specific knowledge bases, and a support ticket agent that assists support engineers. INO consistently improves results over baselines in terms of discoverability and usage of factual corrections, in automated and human evaluations.