Yuzheng Sun

*production models

STAR is good but not enough for a Sr+ role. Instead, you should use H-STAR-G: Hook, situation, task, action, result, and generalization. As a senior candidate, you shall be familiar with STAR already so I won't waste time there. "Hook" is about why we got into the situation. Why? And why you? Why nobody else? It creates tension and brings necessity, so when you solve the problem, your value is obvious. "Generalization" is about taking your solution to the next level. Did the problem reflect a deeper problem? Could the solution apply across orgs? Did you build a tool for others to solve similar problems? Did the company see benefits beyond this one-time fix? The H and G are what bring a sr to staff and principal level.

I have interviewed over 100 data scientists. One of the biggest mistakes they make is going to the "how's" too early. How's are important. Knowing the how's shows we are experts and reliable. But going there too early is a big mistake because it doesn't reflect your value, your thoughts, and your initiative. You rely on the interviewer to guess all of those. Namely, even if you did an impressive task with impressive skills, could someone else do it better? What's your role in this entire project? Was it a team project? Did the manager assign it to you? Did you overcome some challenges? Did you create unique value? Don't make the interviewer guess the answer to these important questions. Show them by telling the why's first.

In case you haven't learned this: Product models are not models you learned at school/Kaggle/online. Anyone in a DS master program can replicate any fancy models, but only a few people can deploy a good production model end-to-end. Why? The challenges are not in knowing which Python library to use, the challenge is in data, engineering, the dynamic environment, performance evaluation, and 50 other steps that are not taught. Build on the expectation that this is not about modelling. This is about the production and the craft. Respect that and you will have a higher chance of succeed in this field.

A conclusion that's true 100% of the time is guaranteed to be useless. Such conclusions are called truisms. In a business environment, we drive insights to make tradeoffs. There is no such thing as making it better for everyone. Such ideals only exist in schools. Embrace that, and identify what's useful, instead of what's "right". There is a saying -- "All models are wrong, some are useful". We can extend it and say "all data scientists are wrong, some are useful." Be that useful data scientist! I have a video to discuss this concept: https://youtu.be/V_5x61AWxwo

Like self-discipline or self-help books, continuous learning is a dangerous trap. To many of us, learning is additive. It's like the hedonist treadmill. We learn for the sake of learning. But is it necessary for you to learn how to write C++? Probably not. The real learning shall be challenge-oriented and action-driven.