Roles by discipline
ML roles with something already in production
The distance between a good applied science role and a frustrating one is measured in deployments, not in how interesting the problem is. There are teams with a repeatable path from notebook to serving traffic, and teams that cannot explain how the most recent deployment got there. Both describe themselves identically in a job ad. So the first thing we ask for is the count of models currently serving traffic, and the story of the most recent one that shipped.
What we ask before we post one
The questions the ad never answers.
- How many models are serving traffic right now, and who keeps them there?
- What did the path from notebook to production look like for the last one, and how long did it take?
- Who owns data access, and how long does an approval take?
- What compute is available for training, and who queues for it?
- Is the team allowed to publish or present externally?
- When a model degrades, who notices first?
What we will not post
We do not post applied science roles where nothing has reached production and there is no funded plan for it to.
For hiring teams
Hiring for this discipline?
We work through the role and hiring criteria with your team as part of embedded recruitment. Talk to Jonathan about what you need to hire.