Search relevance consulting that users can feel.
When people search and don't find, they leave. We find out why results miss, then fix the query understanding, retrieval and ranking behind them, with metrics that show what changed.
Sound familiar?
- Users search, get poor or zero results, and rephrase again and again.
- Search converts worse than browsing, and nobody can say exactly why.
- Relevance lives in hand-written boosts and synonym lists that no one fully understands any more.
- You can't tell whether the last release made search better or worse.
- You added vector search for meaning, and lost the precision keyword search used to give you.
How we fix it.
Relevance audit
Query log analysis, failure taxonomies and side-by-side reviews that show where search breaks and what each fix is worth.
Query understanding
Spelling correction, intent and category classification, attribute extraction and query rewriting.
Hybrid retrieval
Keyword (BM25) and dense embedding retrieval combined so you keep precision and gain recall.
Learning to rank
Ranking models trained on engagement and judgment data, from gradient-boosted trees to cross-encoder rerankers.
Search evaluation
Judgment sets, offline metrics such as nDCG, MRR and recall, and A/B tests that tell you what actually moved.
Deliverables your team keeps.
Common questions.
Do you work with Elasticsearch, OpenSearch, Solr, Vespa or vector databases?
Yes. The methods are platform-agnostic. We work inside your existing stack and only recommend a migration when the evaluation shows it pays off.
Is this only for ecommerce search?
No. Our founder's experience spans ecommerce search at Walmart and Shopify and content search at Pinterest and Roblox. The same methods apply to marketplaces, media, SaaS and internal enterprise search.
Do you need access to our production data?
We need enough access to understand your query logs, catalog or content, and current ranking. We're happy to work under NDA and within your security requirements.
How long does an engagement take?
Most engagements start with a short diagnostic phase. The full timeline depends on your data and stack, and we agree it with you before work starts.
Often combined with.
RAG & retrieval
Make LLM answers right by fixing what they retrieve: chunking, embeddings, hybrid search, reranking and RAG evals.
Learn moreLLM pre-training & fine-tuning
Make an open model fluent in your domain with DAPT, SFT and DPO, from data curation to a model you can ship.
Learn moreAI & search evaluation
Know whether a change helped: offline metrics, LLM-as-judge pipelines and A/B tests your team can trust.
Learn moreHave a search or LLM problem worth solving?
Tell us what you're building. You'll talk directly with the founder, not a sales team. If we're not the right fit, we'll say so and point you to someone who is.