Interviews & narratives¶
15 hours · cheap to prepare, expensive to fail
The smallest pillar and the highest return per hour. Behavioural rounds don't usually win an offer, but they routinely lose one — and unlike everything else in this roadmap, the material is finite. Fifteen hours genuinely covers it.
The distinctive advantage for anyone arriving from production work is that the stories are already real. The failure is arriving with them un-rehearsed and improvising a rambling version of something that was actually impressive.
Modules¶
1 · Stories, written before they're needed · ~5h¶
Eight to ten STAR stories, written out, covering the standard axes: a technical decision you got right · one you got wrong · conflict with a colleague · influencing without authority · a deadline you missed · ambiguity you resolved · something you shipped end to end · something you killed.
Two rules that do most of the work:
- Numbers, or it didn't happen. "Cut latency 60%" lands; "improved performance" doesn't. Dig the real figures out before you need them.
- One story, several angles. The same project can answer conflict, ambiguity and technical judgement depending on which part you foreground. Ten stories covering thirty questions beats thirty thin ones.
Done when: each story is written, timed at 2–3 minutes spoken, and you can tell any of them without narrating the entire project history first.
2 · Design narratives from shipped work · ~4h¶
The bridge between this pillar and AI system design, and the part most people skip.
For each significant system you've built: what constrained it, what you tried that failed, what you rejected and why, what you'd do differently now, and the number that justified the decision. Interviewers reward "we tried X, it failed because Y, so we did Z" far above a clean diagram — but that story has to be prepared, because the scars are the first thing that disappears under pressure.
Done when: three systems are written up this way, and each has at least one honest "what I'd do differently."
3 · Mock interviews with real humans · ~5h¶
The only irreplaceable item in this pillar. Solving silently trains a different skill than solving out loud for a skeptical stranger, and the gap between them is larger than anyone expects until they've been in the room.
Weekly through Phase 3, mixed across coding, system design and behavioural. Peers are free and honest; paid platforms are worth it specifically for a stranger's calibration, which friends can't give you. Record at least one and watch it back — deeply unpleasant, and the fastest correction available.
Done when: you've done six mocks, and the feedback has stopped repeating itself.
4 · Negotiation · ~1h¶
The highest hourly-rate work in this entire roadmap, and it is not optional.
An hour of preparation here can move compensation more than the other 299 hours combined. Know the band before the first call. Never give the first number. Never accept on the phone. Understand what's actually negotiable — base, equity, sign-on, start date, level — and that level is usually the one that matters most, because it compounds.
Done when: you have a researched band for your target roles and a rehearsed answer for "what are your salary expectations?" that doesn't name a number.
The meta-skill: reading rejections¶
Track which stage rejects you. It's the highest-signal data you'll generate, and it's free.
| Rejected at | What it actually means |
|---|---|
| Resume screen | Positioning problem — the resume doesn't read as the role |
| Recruiter call | Story problem — you can't say what you do in 30 seconds |
| Coding round | DSA problem, and only that |
| System design | Design problem — usually structure, not knowledge. If it was the AI round, that pillar |
| Behavioural | Preparation problem, and the cheapest of all of these to fix |
| Final, no offer | Often levelling or budget, and frequently not about you |
Redirect hours toward whichever line keeps repeating. Guessing where you're weak is consistently wrong; the pattern isn't.
What I'm skipping¶
- Interview question mega-lists. Low signal, enormous volume, and they crowd out the practice that works.
- Rehearsing scripts word for word. Reads as rehearsed, and collapses on a follow-up. Learn the beats, not the lines.
- Company-specific trivia. Values pages and leadership principles matter at exactly one company. Prepare that one when it's on the schedule.