{"company":{"name":"Whoop","slug":"whoop","website":"https://whoop.com","category":"hardware"},"question":"What has Whoop shipped recently?","answer":"In the last 30 days, Whoop shipped 4 tracked updates. The most recent was \"CDC at WHOOP: Self-Service Replication for Hundreds of Postgres Tables\" on 2026-09-30.","window":{"days":30,"updateCount":4,"returned":4},"generatedAt":"2026-10-08T07:42:13.903Z","updates":[{"title":"CDC at WHOOP: Self-Service Replication for Hundreds of Postgres Tables","summary":"WHOOP replaced a Debezium-to-Spark pipeline with a two-layer Iceberg-based CDC system using Flink, splitting streaming (Bronze) from upserts (Silver). Service teams now self-manage table replication via Liquibase migrations, with schema changes propagating automatically to downstream systems like Snowflake.","date":"2026-09-30","dateIsEstimated":true,"signalType":"technical","signalTypeLabel":"Technical","sourceUrl":"https://engineering.prod.whoop.com/cdc-at-whoop/","publisher":"engineering.prod.whoop.com"},{"title":"CDC at WHOOP: Self-Service Replication for Hundreds of Postgres Tables","summary":"Whoop replaced a Debezium-to-Spark pipeline with a two-layer Iceberg-based system (Bronze for raw CDC events, Silver for upserted replicas) to enable self-service table replication. Service teams now add tables via GitHub migrations, and schema changes propagate automatically without manual intervention.","date":"2026-09-26","dateIsEstimated":false,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://engineering.whoop.com/cdc-at-whoop","publisher":"engineering.whoop.com"},{"title":"Scaling an ML Inference Pipeline for Batch Workloads","summary":"WHOOP reduced ML inference time for 15.8M tasks from over two months to six days by eliminating HTTP latency, optimizing worker parallelism, and refactoring task chains. Key changes included in-process model execution, CPU allocation tuning, and handling SQS duplicate deliveries.","date":"2026-09-24","dateIsEstimated":true,"signalType":"technical","signalTypeLabel":"Technical","sourceUrl":"https://engineering.prod.whoop.com/scaling-ml-inference-pipeline/","publisher":"engineering.prod.whoop.com"},{"title":"Scaling an ML Inference Pipeline for Batch Workloads","summary":"Whoop reduced ML inference time for 15.8M tasks from over two months to six days by eliminating HTTP network calls, using process spawning to avoid deadlocks, and optimizing SQS message handling. The changes cut per-task latency from 45 to 17 seconds and enabled faster research workloads.","date":"2026-09-23","dateIsEstimated":false,"signalType":null,"signalTypeLabel":null,"sourceUrl":"https://engineering.whoop.com/scaling-ml-inference-pipeline","publisher":"engineering.whoop.com"}],"attribution":{"source":"Spyingbee","url":"https://spyingbee.com/updates/whoop","citation":"Spyingbee, \"Whoop updates\", https://spyingbee.com/updates/whoop (retrieved 2026-10-08)"}}