Productionalizing a Data Science Team
summary
Process recommendations come from running two teams that grow from solo to 4 and from 2 to 10 and proving what scales: two-week sprints with daily standups and sprint planning, JIRA for tracking, repo scaffolding via cookiecutter/yeoman and a shared utilities repo called Hermione's handbag, enforced linters/formatters like Black and CI in Jenkins, REST APIs built with Flask, load testing with locust, checklisted PRs and unit tests, and documentation that lets an intern deploy a model by copying an existing project.
AI-written summary · about
chapters (10)
- 0:00Productionalizing intro
- 2:26Hiring collaborators
- 3:27Agile workflows
- 7:17Repo layouts
- 10:50Coding standards
- 15:00Code reviews matter
- 19:31Model deployment APIs
- 22:59Documentation backbone
- 26:51Recap evolving processes
- 27:43Q&A
related talks
23mData science as a team sport
22mBuild Smarter, Don't Just Build
54mThe secret sauce of data science management
16mRedefining Agile for Data Science: A Team First Approach
28mMaking do: Scaling expertise on teams you can't grow
48mSoftware Teaming: Onboarding, Remote Work, Productivity — Four Years in an Team's Life
20mGetting Data Done with a Pragmatic Data Team
18mThe 'I' in Team: Peer-to-Peer Best Practices for Growing Data Science Teams