I had a 30-min chat today with Stefan Mai, who's been at Amazon and Meta for 10 years.

Here are some questions I asked and his response (poorly paraphrased from what I can recall).

How to pivot from data analytics to SWE?

Frame your resume, tell the right stories. If you collected data, built pipelines, build a product, shipped something, write it all down in your resume.

What makes a candidate stand out? What do you look for when you hire?

A lot of candidates look the same in coding tests. We look more at what they do outside of class, open source projects. There are some people who just have the love for building something, creating products that makes then stand out. These are the people that have deep knowledge about a certain thing, and having a conversation with them you can get signals that they've done a lot of shit.

What are some common mistakes candidates make?

Nerves during interviews. Giving up on a problem at 0, and not trying to solve till 50% or 80%. Not asking edge cases when starting out in interview, which brings up a lot of issues later in your code. Advice is to just take a deep breath before each question. Calm down.

Were there key decisions or actions you took early in your career that contributed most to your success?

It was small changes that eventually led me to where I am. I started out in consulting doing SWE, and I wanted to get into ML. I wanted to get into big tech to learn from great people. Just be sure rate of learning is positive. The industry is fast-paced and you have to be learning constantly.

Thoughts on Gen AI and how it impacts career?

in every hype cycle of AI and ML people make a lot of predictions about what's going to happen, but things end up being the same. Things are definitely going to be different, it's going to be like the internet boom where a lot of business started making websites, APIs, etc. it'll happen with Gen AI as well. For ML, there's less focus on the data acquisition and training models, more on fine-tuning. For SWE, CRUD apps in webdev are going to go away, you can build basic websites with english now, although there are complexities with web dev now. So focus on the fundamentals and just pursue what you're interested in.

Culture difference in Amazon vs Meta?

Amazon operates on a small margin, and it's very process-based. They're focused on how to optimize and improve their operations and processes. It also has a hierarchy where each team is siloed and you don't get access to what other teams are doing. They do have an annual ML conference where it's like academia, they have posters presenting their research. Meta operates on a high margin, they have more money to burn on new things. It's also very unstructured, you won't find documentation for something you need, so you'll be going across teams to get answers. Meta also values emotional intelligence. It's a big company, so experience is very team dependent.

What would you tell your younger self before starting a company?

To start sooner. I wanted to start a company when I was your age, and wanted to get some experience from big tech. For some time I thought I didn't have enough experience yet or wasn't a good enough leader to start my own company. But now I believe 2-3 years is enough at a big tech and it can teach almost everything you need to know to found your own company. But working at big tech for so long did give me the financial ability to pursue my ventures now.

Some mistakes:

• felt intimidating talking to him right off the bat, I could tell he was really intelligent and articulate, everything he said was dense and rich in information.
• I felt like I was asking really cliche and basic questions, and was thinking about better questions to ask midway, which made me lose focus about he was saying.
• I was saying "that's a good answer/point", "thank you for the insights", but had nothing to respond back to or add on to what he said. I was also optimizing to ask as many questions as possible, but I need a less awkward way to respond back.
• I was not confident in asking questions, I was hesitant and incoherent instead of being direct and clear.

Takeaways:

• Treat coffee chats as a conversation with a friend.
• I should follow up on their response so I can latch on to something, give it time to digest and form structure.
• I should do more research and ask questions that are high-value and specific to the person, my questions were too generic.

I found this collection of interviews with ML practitioners on ApplyingML with some good questions and might be using some of these for my future coffee chats.