Isabella Tromba
Software Engineer on Slack’s Search & Learning Infrastructure (SLI) team. Co-authored the Search at Slack engineering post describing Slack’s two-stage LTR system for enterprise message search.
Articles
- Search at Slack — two-stage LTR, work graph signals, position bias oversampling, +27% P@1
Key Contributions
- Position bias fix: oversample clicks from lower positions to equalize positional distribution in training data
- Work graph as personalization signal: user affinity to message author, channel priority
- Two-stage architecture: Solr first-pass + application-layer re-ranking with SparkML SVM