# Craft

> Enterprise search, relevance, and the leadership of engineers.

## Enterprise search

Ten years centered on Apache Lucene and Apache Solr, with production work on Elasticsearch, OpenSearch, Milvus, and Lucidworks Fusion.

- Schema design, analyzers, and BM25
- Sharding, replication, and SolrCloud
- Autosuggest and faceting
- Java, with Python, SQL, and standard cloud tooling beside it

Subjects: Enterprise search, Apache Lucene, Apache Solr, Elasticsearch, OpenSearch.

- [Public search at country scale](https://oppority.com/work/public-search.html)
- [One catalog, many storefronts](https://oppority.com/work/retail-search.html)
- [Concept search before vectors were the default](https://oppority.com/work/concept-search.html)
- [Return one record, not a list](https://oppority.com/work/one-best-record.html)

## Relevance and migrations

A migration is complete when the new system returns a result a person can trust, not merely when the cluster is healthy.

- Similarity chosen for the field
- Overlap on the top results when a platform changes
- Like-for-like checks when the product returns one record

Subjects: Relevance, Apache Solr, OpenSearch.

- [A move that kept the answers](https://oppority.com/work/platform-move.html)

## Leading engineers

I manage a search engineering team and still work hands-on. The useful part of that mix is translation: a need, a relevance behavior, and a system other engineers can operate.

- A need, stated so an engineer can act on it
- A relevance behavior the team can explain
- A system that runs without one person

Subjects: Engineering management.

- [The team's output](https://oppority.com/work/search-team.html)
- [Your First 90 Days as a New Engineering Manager](https://oppority.com/books/first-90-days-engineering-manager.html)

## Current study

Retrieval in the current wave of systems, and how an answer engine reads a page, structures context, and attributes a citation.

- Quantization, HNSW, and when exact similarity beats approximate search
- Hybrid BM25 plus dense retrieval, and how to check the blend
- Luxir, a new open-source hybrid engine, read by running it

Subjects: Hybrid search, Answer engines, Luxir.

- [A new engine, tried on a laptop](https://oppority.com/work/reading-luxir.html)
- [Run Luxir on a Mac or Windows laptop](https://oppority.com/writing/run-luxir-on-a-mac-or-windows-laptop.html)