From Heuristics to Learned Ranking: Airbnb Search's Evolution
How Airbnb's search ranking moved from hand-tuned heuristics to gradient-boosted trees and eventually neural networks, and what each transition actually bought.
Engineering at scale
11 articles · 43 min of reading
How Airbnb's search ranking moved from hand-tuned heuristics to gradient-boosted trees and eventually neural networks, and what each transition actually bought.
Ranking gets the attention, but Airbnb's search infrastructure has to first find every listing that could plausibly match a map, dates, and filters, fast.
How Airbnb replaced inconsistent, hand-rolled UI across its apps with a shared design language and component library engineers and designers could both trust.
How Airbnb built a payments platform that handles dozens of currencies, split payments between guests, and reliable payouts to hosts around the world.
Why Airbnb kept getting conflicting numbers for the same metric across teams, and how Minerva's centralized definitions and certification process fixed it.
Why Airbnb's best analyses kept getting lost in personal notebooks, and how Knowledge Repo made data science writing peer-reviewed, searchable and reusable.
Before service meshes existed, Airbnb built SmartStack — Nerve and Synapse plus local HAProxy — to make service discovery reliable during its microservices split.
Airbnb's honest account of splitting its Rails monolith into services, the sprawl it caused, and the platform investment that fixed it.
Airbnb adopted React Native to unify mobile development across iOS and Android, then spent two years learning why it decided to sunset the experiment.
Why cron and ad hoc scripts stopped working for Airbnb's data pipelines, and how the internal tool built to fix it became the industry-standard workflow scheduler.
How a weekend hackathon tool for exploring data at Airbnb grew into Apache Superset, one of the most widely used open source BI platforms.