Data Renegades Ep. #7, Truth-Seeking Data Systems with Bryan Bischof
15 snips
Feb 3, 2026 Bryan Bischof, a PhD mathematician turned AI and ML builder, has built recommendation and streaming systems for startups and taught data science. He discusses building terabyte-scale streaming inference, surprising predictive features, debugging recommendations gone wrong, experiment design to settle disputes, and why truth-seeking should guide data teams.
AI Snips
Chapters
Books
Transcript
Episode notes
From Pure Math To Terabyte Streaming
- Bryan Bischof left a math PhD to join industry after meeting applied researchers and built terabyte-scale streaming inference systems.
- He learned Scala and Spark while solving real-time time series problems at his first startup job.
Automating P&L And Coffee Recommender
- At Blue Bottle Coffee Bryan automated the P&L and built a forecasting engine within weeks to optimize cafe operations.
- He ran experiments like pricing and availability and used a survey-driven random forest to power a coffee recommender.
Too Many Backpacks Bug
- At Stitch Fix Bryan shipped a "steal this look" recommender that bizarrely returned many backpacks due to data quirks.
- The bug arose from multi-category backpacks and L2 distance placing backpacks unusually close to everything in latent space.
