Case Studies
Field reports on how real teams built, scaled, and fixed search in production — with the constraints and dead-ends textbooks omit. See also the curated MOC - Case Studies.
- Airbnb - ML-Powered Experiences Ranking — ML ranking and the data flywheel
- Canva - Search Pipeline Modernization — modernizing a legacy search pipeline
- Etsy - Search Quality and Query Understanding — query understanding in a long-tail marketplace
- foodpanda - Classifying 300K Noisy Search Terms Across 16 Markets — multilingual query classification and the 37% “Others” bucket
- Hybrid Fusion Failure - BM25 Displacing Reference Documents — a cautionary failure: unnormalized score fusion silently dropping the right document
- Kleinanzeigen - Vespa Migration for Homepage Feed — migrating to Vespa for feed ranking
- Netflix - Content Search Architecture — search architecture for a content catalog
- Reddit - Vector Database Selection — how to actually choose a vector database
- Skyscanner - Learning to Rank for Flights — Learning to Rank in production
- Slack - Enterprise Message Search with LTR — enterprise/personal search with LTR
- Uber Eats - Scaling Search for Food Delivery — geo-constrained retrieval and ranking at scale
- Vespa - Ranking Without Labels on CORD-19 — four stages from tuned BM25 to a synthetically trained cross-encoder, on a corpus with no labels
- Vinted - Migrating Search from Elasticsearch to Vespa — a billion-item search platform’s ES→Vespa migration
- Zalando - Self-DoS via Facet Aggregation — a cautionary failure: facets that took search down