MLOps for the rest of us

We all heard of the large tech companies running hundreds of ML models in productions and having impressive setups. However, most of us work at much smaller companies. Do you run more than 30 models?

Should we copy what the organizations with thousands of developers are doing? Isn't it overkill?

Let's talk about MLOps at a mid-size startup. What should you do when you are the only MLOps engineer in the company? How can one person keep 15 models running without becoming a bottleneck for the machine learning engineers training the models? How can we automate testing the models, deployments, and monitoring?

How can we do all of that, end the workday at 5 pm, and don't worry about production failures at night or on weekends?

Does it sound too good to be true? Let me also tell you about all the mistakes I have made and the "joy" of Sunday morning debugging.