Official account of the Airbnb Technology Team, sharing Airbnb's tech innovations and real-world implementations, building a world where home is everywhere through technology.
The article shows how Trio stores navigation state in a ViewModel‑backed stack, uses type‑safe routers for screen and activity transitions, supports async save‑and‑exit, result‑returning navigation, and interoperates with legacy fragments, enabling modular, testable navigation in Jetpack Compose.
Airbnb’s Android team introduced Trio, a Compose‑only framework built atop the Mavericks state‑management library that replaces Fragment‑based navigation with type‑safe, modular page routing, ViewModel‑driven state, compile‑time validation, and dedicated developer tools, now powering the majority of the app’s screens.
Airbnb’s Android Page Performance Score (PPS) framework instruments fragments to collect user‑centric metrics such as TTFL, TTIL, MTH, ALT and RCLT, using a standardized logging config, LoadableView interface, and visibility algorithms, enabling detailed performance analysis and automated alerts for mobile teams.
Airbnb’s iOS Page Performance Score implementation introduces a PPSStateMachine that ties a UIViewController’s lifecycle to metric collection—tracking first layout, initial load, scroll thread hangs, additional and rich content load times—using nanosecond timestamps, state‑machine protocols, and view‑responder discovery to emit standardized performance logs.
Airbnb tracks five real‑user performance metrics—Time To First Contentful Paint, Time To First Meaningful Paint, First Input Delay, Total Blocking Time, and Cumulative Layout Shift—using a mix of browser APIs and custom polyfills, combines them into a weighted Page Performance Score, and leverages that score to guide trade‑off decisions and detect regressions.
Airbnb created the Page Performance Score (PPS), a unified 0‑100 metric that aggregates platform‑specific initial‑load and post‑load user‑centric measurements for Web, iOS, and Android, using weighted curves to enable cross‑page, cross‑team comparisons, track organizational weighted averages, and evolve with new metrics while preserving a stable scale.
Airbnb’s Attribute Prioritization System uses a machine‑learning pipeline called LATEX to extract and map guest‑mentioned amenities, activities and places from reviews, messages and tickets, then predicts and ranks the most important attributes per listing, giving hosts personalized suggestions to improve listings and match traveler needs.
Airbnb’s Riverbed framework unifies streaming CDC events and batch Spark jobs behind a GraphQL‑based declarative API to automatically build and maintain distributed materialized views, using Kafka‑partitioned ordering and version control to deliver billions of daily updates with low‑latency reads for features such as payments and search.
Airbnb’s Listing Attribute Extraction Platform (LAEP) uses a custom NER model, word‑embedding mapping, and a BERT‑based scorer to automatically pull, normalize, and validate structured attributes from hosts’ unstructured text, boosting coverage for downstream tools and enhancing guest‑host matching at scale.
Airbnb’s new computer‑vision pipeline trains a deep‑learning aesthetic model with an EMD loss to rank photos, automatically sorts new‑listing images by design and room type, and scales real‑time similarity search via HNSW‑based ANN on AWS OpenSearch, boosting click‑through, bookings, and enabling unsupervised visual recommendations.
