QREF: Optimizing DeepSearch Agents via Query Refinement and Evidence Filtering
October 16, 2026
QREF uses supervised fine-tuning and reinforcement learning to align search queries with reasoning intent. The system implements a plug-and-play query refiner and an evidence filter to reduce retrieval noise and improve coverage in complex information-seeking tasks.
HOW THIS AFFECTS YOU
●
builderYou can use these methods to reduce redundant API calls and improve the signal-to-noise ratio in agentic search loops.
●
researcherThis provides a framework for optimizing agent-environment interactions through RL-driven query generation.