Aggressively Helpful Platform Teams // Stefan Krawczyk // MLOps Coffee Sessions #49
MLOps.community
Stitch Fix's approach to model deployment and data science autonomy is the focus, featuring Stefan Krawczyk, a manager on their data platform team. Krawczyk details Stitch Fix's "full stack data scientist" model, where individuals are empowered to solve problems end-to-end, enabled by a first-class experimentation system. The discussion covers how Stitch Fix manages numerous independently developed models through experimentation, accountability, and organizational maturity, and also addresses challenges encountered, such as resource overuse, emphasizing a culture of responsibility. Krawczyk introduces Hamilton, a micro framework for creating data frames, and Model Envelope, Stitch Fix's system for streamlining model productionization, emphasizing its ease of integration with internal tooling and its model-agnostic design.
Part 1: Introduction, Background
Part 2: Culture, Autonomy, Strategy
Part 3: Platform Structure, Hamilton Framework
Part 4: Model Envelope, Technical Implementation
Part 5: Scaling, Serialization, Success Metrics
Sign in to continue reading, translating and more.
Open full episode in Podwise
