ML interview prep — the focused 30-day plan

Most ML interviews have:

1. Coding (DSA) — 1-2 rounds. Leetcode-style.
2. ML breadth — 1 round. Concepts, trade-offs, math at the level you'd ace whiteboard.
3. ML depth / system design — 1 round. Design a recommender, RAG, classifier pipeline.
4. Behavioural — 1 round. STAR-format stories.

Mix varies by company:
- Big tech: heavy DSA + ML system design
- Startups: more code + ship-it-on-a-laptop demo
- Research-leaning: depth + a 30-min paper grilling

Know the format before you prep.

Days 1-7: DSA. The 8 patterns from week 4. 2 problems/day.

Days 8-14: ML breadth. Hands-On ML chapters; for each, write a 1-page summary + an honest 'I don't get this part' note.

Days 15-21: ML system design. 'How would you build X?' — pick from real prompts (RAG, ranking, fraud, ETA). Write your design before reading examples.

Days 22-28: Mock interviews. interviewing.io, friends, Discord. Real-time pressure changes what you 'know'.

Days 29-30: STAR stories — 5 prepared. About impact, conflict, failure, ambiguity, leadership.

Same skeleton, every system design question. Adapt boxes to the problem.

User → API → Feature pipeline → Model → Output cache → Logging
                              ↓
                    Offline training pipeline → Eval → Model registry → Deploy

Draw it on the whiteboard. Walk through each box. Talk trade-offs at each step. The interviewer is grading your structure, not your perfect answer.

ML interviewers grade thinking, not just answers.

My rule: 80% of solving a problem out loud, 20% silent.

- Restate the problem before solving
- Talk through brute force; state Big-O
- Pick the pattern out loud
- Code, narrating each line
- Test, narrating edge cases

Silent solving costs you. Even if you reach the right answer, the interviewer didn't see how. Show the process.

Day 88 done.

- 4 rounds, know the mix
- 30-day plan, 1hr/day
- ML system design canvas
- Talk 80% of the time

Tomorrow (Day 89): the learning loop that keeps you sharp after the interview.