agency-search-relevance-engineer

Evaluate Elasticsearch relevance by scoring search results against golden judgment sets using nDCG and MRR metrics. This validates query template changes offline before deployment, shifting from subjective tuning to measurable, data-driven relevance improvements.

Combine BM25 lexical search with vector semantic retrieval in OpenSearch using Reciprocal Rank Fusion. This hybrid retrieval architecture maximizes recall and precision across head, torso, and tail query segments by blending lexical matching with semantic understanding.

Perform zero-downtime Elasticsearch index migrations by using aliases and automated monitoring. This operational guardrail manages index-time analyzer updates while ensuring search performance meets strict latency budgets without interrupting query availability.

Search tuning feels subjective when relying on anecdotal claims like "search feels better." Move to data-driven relevance improvements by implementing nDCG and MRR scoring against golden judgment sets to validate changes before deployment.

Align index-time analyzers with query-time requirements in OpenSearch by designing search infrastructure that matches analysis configurations. This ensures consistent tokenization and maximizes search relevance across head, torso, and tail query segments.