Elasticsearch and OpenSearch relevance consulting
Your engine is probably fine. The relevance on top of it usually isn't. I tune analyzers, fields, BM25, hybrid search and ranking in the Elasticsearch or OpenSearch cluster you already run, and measure every change.
A good fit if…
- Results are wrong, and the fix so far has been more boosts and synonyms.
- Exact matches on SKUs, part numbers or names fail because of how text is analysed.
- You added kNN or vector search and aren't sure how to combine it with BM25, or whether it helped.
- You're considering a migration to fix relevance, and want to know if it would.
Look elsewhere if…
- Your cluster is falling over: you need performance and operations help first.
- You want an Elastic licence or support contract. I'm independent of the vendors.
Deliverables your team keeps.
Vector and hybrid search in Elasticsearch and OpenSearch.
Both engines now do vector search well: dense_vector and kNN in Elasticsearch, knn_vector in OpenSearch, both on HNSW indexes. The hard part isn't turning it on. It's combining it with BM25 so that each covers the other's blind spots, without losing what keyword search already did well.
The defaults are where it usually goes wrong. Elasticsearch's rrf retriever and OpenSearch's hybrid query with RRF merge two separately ranked lists. That's a safe start, but it can quietly bury exact matches: in a benchmark I ran on Wayfair's product search data, a query for a specific rug had the right product at #1 in BM25 and at #56 after RRF.
In practice, I check whether kNN is finding what it should, compare RRF with a tuned weighted combination (Elasticsearch's linear retriever or OpenSearch's normalization processor), set weights by query type, and add learning to rank or a reranker where it pays off. Every step is measured per query type, with the same method as my public benchmark.
From brief to handover.
Review mappings, analyzers and queries. Sample real queries and label what should come back.
Baseline the metrics and find which failures come from analysis, query structure or ranking.
Fix them in your cluster, in order of impact, with every change measured offline first.
Documented mappings and queries, an evaluation harness, and a team that knows why each setting is there.
Works with what you run.
What I've built before.
How it's priced.
Not sure where to start? Most teams begin with a fixed-scope search relevance audit.
Starting prices; the final quote depends on scope and follows a free intro call.
Remote, working with teams in the US, UK and Europe. Async-first, with live time during your Pacific mornings, or your morning and early afternoon if you're in Europe. Contracted and invoiced in USD through Infersense Labs LLC (Wyoming, USA).
Questions buyers ask.
Do we need to migrate to a vector database?
Usually not. Elasticsearch and OpenSearch both support vector and hybrid search. I recommend a migration only when the evaluation shows it pays off.
Can you work within our managed service (Elastic Cloud, Amazon OpenSearch)?
Yes. The relevance work happens in mappings, queries and ranking, which work the same on managed services.
Do you also do performance tuning?
Where it affects relevance, like ANN recall settings. For cluster sizing and operations, I'll point you to specialists.
Can you help with vector search consulting for Elasticsearch?
Yes. I set up and evaluate kNN and dense vector search in Elasticsearch or OpenSearch, choose and test the embedding model, and combine it with BM25 as hybrid search, measured against your real queries.
Should we use RRF or a linear combination for hybrid search in Elasticsearch?
Start with RRF if you have no labelled queries. With a few hundred, a tuned linear combination often does better, and in my public benchmark, giving every candidate both scores beat merging two lists. Test both on your own queries, per query type.
How do you avoid breaking queries that work today?
Every change is tested against an evaluation set covering your important query types, so wins on one kind of query don't hide losses on another.
Tell me what you're hiring for.
A free 30-minute call with me. You'll leave with a clear next step, and an honest answer on whether I'm the right person for it.