Real Interview Experiences
Learn what to expect directly from candidates and interviewers who've been through it.
- 1,326 experiences
- 314 companies
- 37 roles
- +207 added this month
CohereMachine Learning Engineer · L6“The system design interview was data collection, SFT and RL stages. It went well for me. It was focused on LLM post training techniques. The ML coding was worst for me although I practiced the same question but using PyTorch and doing without PyTorch and without any documentation was challenging for me.”
AnthropicMachine Learning Engineer · L4“The process is kind of cold. All tasks are performed on codesignal. You are basically do the assignment by yourself with the computer and camera on.”
OpenAIResearch Engineer · L5“The problem was challenging. There were many conditions you had to satisfy and some aspects were a bit underspecified. The interviewers were stone faced and gave zero commentary. I ended up making a small bug which I wasted about 15 minutes trying to fix.”
GoogleMachine Learning Engineer · L6“Interviewer was very comforting but focus on more Eval part. Prepare eval better along with failover strategy, model choices.”
MicrosoftMachine Learning Engineer · L3“What went well: The live coding and system architecture rounds went smoothly. I was able to articulate my design choices clearly, especially around API design, database strategy, and maintaining high availability for microservices. What didn’t go well: The initial behavioral portion felt slightly rushed because we spent extra time discussing specific cross-functional project dynamics from my previous roles. How does this interview compare to other interviews you've done? It was significantly more interactive. Rather than giving static answers, the interviewers actively challenged my assumptions and asked "what if" scenarios, making the evaluation feel comprehensive and realistic to day-to-day work.”
AppleMachine Learning Engineer · Intern“It was a good interview overall, I was just not prepared for it. Because it was a researcher role, i was first ask an my experience doing AI/ML research then we moved on to the technical problem, that I completely flunked because of nervousness and totally forgoing reviewing DS and Algorithms”
WorkdayMachine Learning Engineer · L6“The process was fairly quick. Recruiter gets back very fast. Good experience. All interviews completed within a week.”
GoogleMachine Learning Engineer · L4“Went harsh and the interviewer was rude while asking the questions it was 45 mins but they extended it beyond 1 hr”
AmazonMachine Learning Engineer · Intern“It started with self-introduction and the interviewer wanted to know more about my research. So we main discussed about the papers and the potential topic I can do during the internship”
MetaMachine Learning Engineer · L4“The overall process was rigorous and comprehensive, more challenging than many interviews I have done with other large technology companies. However, the recruiter’s guidance was accurate and genuinely helpful, which I really appreciated. The interviewers were professional and respectful, and I felt the process was designed to evaluate both technical depth and practical problem-solving ability. What went well was that the process gave me opportunities to discuss my experience, research background, and approach to problem solving. What did not go as well was that a few questions caught me off guard, especially where I needed to connect fundamentals, implementation details, and communication under time pressure. Even though I was not able to answer every question as well as I wanted, the experience helped me understand which areas I should strengthen further.”
Preview full access of other recent interview experiences
AnthropicSoftware Engineer, Safeguards · L5“The weirdest Anthropic round was the company values interview. It was almost like a therapy session, and honestly if you went to a therapist at some point, you will pass that round much more easily.”
NetflixEngineering Manager, Tooling“What was very unusual is they didn’t give me any tooling to draw the system design, so I just sketched it on a piece of paper and talked them through it, then we got into this oddly deep debate about whether hover-over history should count as a recommendation signal.”