Technical Deep Dive (slide-based)

Prepare slides about a past project; interviewer drills the technical detail. Research-org loop always has it; bar is high. Requirements Prepare slides about a technical project you led and can defend in depth. Cover the architecture, key technical decisions, trade-offs, and measurable impact. Expect detailed follow-ups on implementation choices, what you would change if starting again, and where AI could improve the system even when the original project was not AI-based. Notes One candidate left the round sensing the interviewer thought project depth wasn't enough — the read on the call matched the eventual reject. A canonical deep-dive flame-out: too much content, ran out of time, interviewer asked 'why use hashing for embedding in your search system instead of BERT' (candidate's internal reaction: 'how do you embed an item id with BERT?') — deep-dive bomb killed the whole loop. Deep dive + HM BQ only unlock after the technical rounds pass — reaching them is itself a signal. Interviewers see many strong projects — the bar is high. Preparation Slides should not be overstuffed; deep-dive 1-2 strongest projects. Anticipate 'why not X' and 'how does this scale 10x'. Structure: motivation → tech tradeoffs → metric improvement → lessons learned. Don't fake expertise in areas you don't own.