arxivcs.CLcs.AI2026-07-05
Risk-Constrained Freshness-Aware Semantic Caching for Open-Web Retrieval-Augmented LLMs
Muhammad Mansoor, Tahir Ahmad, Yeo-Chan Yoon
Semantic caching reduces the latency and cost of retrieval-augmented generation (RAG) by serving cached answers to semantically similar queries, but most existing methods do not model the time-varying freshness of open-web evidence. We present FreshCache, a three-tier semantic ca…