Johann Schleier Smith, Co Founder and CTO, ifwe
summary
Company if(we) describes an event history architecture and Scala-based runtime that turns streams into live features, avoiding SQL-to-Java rewrites and supporting >1,000 updates/sec against ~10M candidates. The demo uses counters and a TF‑IDF feature, trains a logistic regression on a Best Buy Kaggle set, and scores with an 84.4% click hit rate. Iteration time falls to days or weeks, core tech is proprietary and a lighter version is on GitHub.
AI-written summary · about
chapters (7)
- 0:00Agile machine learning
- 2:50Real-time recommendation challenges
- 5:39Event history as data structure
- 10:14Model design with events
- 13:03Live feature engineering demo
- 21:47Rapid ML iteration benefits
- 23:12Key takeaways and open-source
related talks
18mMachine learning with Flink in Weibo
04mLeif Wickland Interview
30mSmart Streams: A Real-time framework for scoring Big and Fast Data
45mReal-time, Streaming Advanced Analytics....
28mReal-Time Revolution: Kickstarting Your Journey in Streaming Data
05mInterview
13mSearch through your data with Weaviate
31mCreating recommender systems with Spark, Scala and Prediction.Io