How much MLOps do I need to know for a "Junior Data Scientist" role in 2023?

All answers to this question.

The days of a Data Scientist just handing over a .py file to an engineer are over. In 2023, knowing the basics of Docker is non-negotiable. You don't need to be a Kubernetes expert, but you should be able to containerize your model so it runs the same way on your laptop as it does on a server. I suggest adding a "GitHub Action" to your project that automatically runs a set of tests on your model every time you push code. This shows you care about "Reproducibility," which is the biggest pain point in corporate Machine Learning right now.

   Answered 2023-11-17 by Sandra Mendez

Sandra, do you think using a managed service like AWS SageMaker or Google Vertex AI is "cheating," or is it actually better to show I can use enterprise-level tools?

   Answered 2023-11-19 by Jeffrey King

• Jeffrey, it's definitely not cheating! In fact, most big companies want you to know those platforms because that's where their production workloads live. If you can show a project where you used SageMaker for hyperparameter tuning and model hosting, you're signaling that you're "ready to work" in a professional cloud environment. Just be sure you can explain what the managed service is doing under the hood so you don't look like you're just clicking buttons.

     Commented 2023-11-20 by Sandra Mendez

Accuracy is vanity, but latency is sanity. A model that is 95% accurate but takes 10 seconds to respond is useless in production. Show you can optimize for speed!

   Answered 2023-11-22 by Maria Lopez

• Exactly, Maria. I’d rather hire someone with an 85% accurate model that responds in 50ms than a 99% accurate one that crashes the server. Engineering constraints matter!

     Commented 2023-11-23 by Ryan Miller