Succeeding at Scale: Enterprise Retrieval Benchmark Construction and Index-Preserving Query Adaptation for Multi-Tenant Search
Researchers have developed DevRev-Search, a new benchmark for passage retrieval in technical customer support, built using an automated pipeline. This benchmark addresses the challenge of adapting retrieval systems in multi-tenant environments where curated relevance labels are scarce and full model re-indexing is impractical. The study also evaluates index-preserving query-only adaptation strategies, demonstrating that parameter-efficient fine-tuning of query encoders offers a strong quality-efficiency trade-off for scalable enterprise search. AI