AI BriefingAirbnb Tech news
The latest AI and developer news from Airbnb Tech, with short summaries.
Visit website →Airbnb open-sources Skipper, an embedded workflow engine for Java and Kotlin
The Apache 2.0 licensed tool allows teams to define durable multi-step processes as Java or Kotlin classes without a separate orchestration cluster.
Airbnb introduces CASTLE framework to solve cold-start problem for natural language search
The framework achieves a KL divergence of 1.01 against real user queries, a 9.2x improvement over previous baselines.
Airbnb Builds System to Capture and Replay Real Database Traffic for Load Testing and Upgrades
The system enabled Airbnb to migrate its MySQL fleet from 5.7 to 8.0 without major production incidents by identifying latency regressions and compatibility issues offline.
Airbnb Introduces Proximity Features for Cold-Start Personalization Without User Identity
The system groups users into buckets of approximately 1,000 based on geo-IP data to generate location-aware features without persistent identifiers.
Airbnb Reduces Recommendation Data Latency from 2 Days to Under 1 Minute by Adding Push Mode and NRT Features to Chronon
Airbnb introduced Push Mode and NRT Model Transform to Chronon, reducing data latency in its recommendation system from 2 days to under 1 minute.
Airbnb Releases 'Insight Miner' AI Agent Harness for Unstructured Data Analysis
Airbnb has released 'Insight Miner', an AI agent harness that improves the reproducibility and speed of unstructured data analysis.
Project Lighthouse Part 3: Introducing project-lighthouse-anonymize
Airbnb has open-sourced the privacy-preserving anonymization code for Project Lighthouse and published related technical papers.
How to Know When the Pandemic Is Over (and What Models Needed to Learn)
Airbnb provided criteria for deciding whether to refit, respecify, or hold models during retraining.
Flexible Authentication: Airbnb Redesigns Authentication for Millions of Users
Airbnb has built Flexible Authentication, which selects authentication methods based on user and context.
Eval-driven development: Lessons from evaluating GenAI at scale
To ensure the reliability of its GenAI products, Airbnb introduced 'Eval-driven development,' treating evaluation not as a simple post-hoc check but as a core engineering process.
Personalizing Airbnb Search Through Guest Journey Learning
Airbnb has built a Transformer-based sequence model that learns guests' long-term behavioral patterns to recommend optimal accommodations.
From Weeks to a Day: How to Speed Up LLM Evaluation to Enable Iterative Improvement
Airbnb shortened its LLM evaluation and experimentation cycle from weeks to just a single day through infrastructure improvements.
Beyond a Single Product: How Airbnb Evolved Its Data Architecture for the Multi-Product Era
This explains how Airbnb redesigned its data architecture to move away from a single-service model and support a diverse portfolio of products.
Sitar-agent: Building a Reliable Dynamic Configuration Sidecar for Large-Scale Environments
Airbnb built **Sitar-agent**, a Kubernetes Sidecar, to reliably deliver dynamic configuration to thousands of service instances.
When History Betrays You, Borrow from Geography
The article body could not be verified due to a Medium system error.
Airbnb Scales Its Identity Graph into a Unified Knowledge Graph Infrastructure
Airbnb migrated its identity graph to a unified knowledge graph built on JanusGraph and DynamoDB.
Viaduct 1.0 and the Future of Airbnb's Data Mesh
Airbnb has unveiled Viaduct 1.0, turning an internal tool into a community project.
Monitoring reliably at large scale
Airbnb broke the monitoring circular dependency using Kubernetes, Envoy, and a Dead Man's Switch.
Building a Fault-Tolerant Metrics Storage System at Airbnb
This covers how Airbnb built a metrics storage system that is resilient to failures.
Privacy-First Connections: How to Strengthen Social Experiences at Airbnb
An article about designing social connections at Airbnb with privacy-first principles.