The candidate presents a past technical project they have worked on — ideally one with high technical complexity and relevance to the team's domain (e.g., AI/LLM systems, distributed systems). The interviewer then conducts a deep-dive Q&A lasting roughly 45–60 minutes, probing the candidate's understanding of why specific decisions were made, what tradeoffs were considered, and how the system was architected. The session is also used as a behavioral/values signal: interviewers ask questions aligned with the company's core values (e.g., cross-team conflict resolution, handling ambiguity). The round is often combined with a 15-minute behavioral/values block.
Common mistakes: Not being able to come up with a concrete example of cross-team conflict resolution — one candidate specifically noted this question caught them off guard even though other behavioral questions were well-prepared.; Weak story preparation for behavioral questions. One candidate noted "the behavioral round did not go well, the stories need more preparation".; Being hung on project deep dive and values without fully understanding why — one candidate said they still felt it was a pity and felt the HM cares a lot about background fit and only engages deeply with projects that are a good fit.
Interviewer hints: HR explicitly provides a detailed prep prompt before the onsite describing what to prepare, including behavioral/values topics. Candidates are told to follow it closely.; The project deep dive is run by the HM, who probes very deeply — one HM spent a full hour on a single project asking about why, how, and tradeoffs.; The HM reportedly cares about domain fit: for an AI agent role, the HM was most engaged when the candidate's project was directly AI agent-related.
What passers do: Candidates who did well chose a technically complex project highly relevant to the team's domain (e.g., AI agent work for an AI agent team) and prepared to explain every design decision, tradeoff, and 'why' in detail.; HR provides a detailed prep prompt before the interview describing what to expect in the behavioral/values block. Candidates who used this prep material to prepare specific stories passed the behavioral portion; those who didn't prepare stories struggled.; The project deep dive portion is approximately 45 minutes, followed by ~15 minutes of core values behavioral questions. Candidates who were aware of this structure and paced accordingly did better.
Alternative approaches: Domain-aligned project selection (Choosing a project closely related to the team's domain (e.g., RAG/LLM for an AI team) generates more interviewer engagement and richer discussion, but may expose gaps if the candidate is less fluent in that space.)