Enabling simple ML ops at Stitch Fix

on YouTube ↗40m

" summary

Stitch Fix builds a configuration-driven ML platform that turns a serialized model plus metadata into deployable services and jobs, used by 145+ data scientists. A model envelope stores function bytes, environment, feature specs and tags; a cron runs every five minutes to compile and deploy microservices (AWS, Docker) or schedule batch jobs via Airflow + Flotilla with Spark and pyarrow map_partitions. Tag-driven selection and tag-based queries plus derived function shape enable automatic validation, monitoring, and a shared metrics dashboard — and it works in production.

AI-written summary · about

" chapters (11)

1. 0:00Intro
2. 1:44Agenda
3. 2:20Takehome
4. 3:12Stitch Fix
5. 5:31What is Data Science
6. 7:02The problem
7. 10:39Build or buy
8. 19:29Online inference
9. 26:44Metrics tracking
10. 30:09Data scientists concerns
11. 31:46Future questions

" related talks

• 32mScaling Personalization: DevOps at Stitch FixRandy Shoup · DevOps Enterprise Summit San Francisco 2017
• 25mPersonalization and Retail: Lessons from Stitch FixBrad Klingenberg · Computational Theories of the Brain 2018
• 34mAugmenting Human Decision Making with Data ScienceKelsey Pedersen · PyCon 2018
• 20mBrad Klingenberg, Director of Styling Algorithms, Stitch FixBrad Klingenberg · MLconf SF 2015
• 26mManaging ML Models @ ScaleSrivathsan Canchi, Tobias Wenzel · OpML '20
• 29mBuilding an ML Platform from ScratchAlon Gubkin · Scale By the Bay 2021
• 30mContainer workflows for data science and MLWilliam Benton · DevConf.CZ 2019
• 26mAI at Stitch FixChris Moody · AI By the Bay 2017