A bit of background first. I spent 6 years at a big fintech, and most of that time went into building an internal experimentation platform. I want to be honest about the niche angle, because I think people assume the opposite of what happened. I never planned to specialize. I sent applications everywhere, senior backend roles, anything that looked decent. Only when I reviewed my own funnel at the very end did I notice that the experimentation roles were the ones turning into interviews. About half of my full loops were experimentation roles, and my only offer came out of one of them.
The numbers, because I logged everything. I searched actively for about a year: around 115 companies, 200 submissions, 13 conversations with a human, 11 full loops, and 1 offer. So only roughly 11% of my applications led to talking with a real person. After that point, I made it to a full loop about 85% of the time.
The Microsoft opening started with a recruiter reaching out to me about a senior role on an experimentation-adjacent team inside CoreAI. It matched what I had been doing for years, so I said yes. The whole loop took about 3 weeks and had 5 rounds.
Hiring manager screen. We did a deep dive on experimentation (sync versus async validation, latency budgets, health checks), and then a live coding problem that started in 2D and had to be generalized to N dimensions.
Traditional coding. First an intervals problem solved with a min-heap, then a BFS on a grid where the state carries extra information. As I read it, they cared about whether I would move from a complicated approach to a simpler one, rather than just getting the first working solution.
AI-assisted coding. This round was a different animal. It was a real repo with multiple files and 5 issues to fix in about 45 minutes, and I got through 3. The hardest was a numerical underflow bug, and the AI agent kept giving me back the same broken code. My read is that the round tests how you think when the AI is wrong, and not how well you write prompts.
System design. It was a rate limiter, but most of the hour was deep dives where I had to actually defend each of my choices.
Behavioral. A full hour, yet only 3 questions, all tied to Microsoft's 7 competencies. There was no retro.
Worth knowing: because this was a senior-level role, I believe coding was deliberately a secondary signal on this team and system design carried the most weight. That is why I put most of my prep there.
What helped:
- An online system design course I worked through. It follows requirements, entities, API, high-level design, then deep dives, and I leaned on that structure heavily for the design round. I realized in December that my system design was weak, so I did as many practice problems as I could after that.
- For the AI round, practice the workflow, not algorithms. Take a real project from start to finish with an agent, and get used to reading and debugging what it gives you while talking out loud, as if the interviewer were there.
- Match each behavioral story to a competency before you go in. I did not memorize them word for word, just an opening sentence and a flexible middle.
- Design from first principles rather than saying "this is how we built it at my job." That also removes any confidentiality worry.
- Negotiate even if you have no competing offer. A structured ask moved my package into the top quartile for the level, and the recruiter handled it well.
The one thing I would tell people: if you have a specialty, find out sooner than I did whether it is already working in your favor. Mine was in front of me for months and I never saw it.
Discussion
Loading comments…