AAI Agentic Handbook
Interview prep

Seven days to an AI engineering loop

About 3–4 focused hours a day. It assumes you can already code; if you're earlier in your career, run the Budding and Mid tracks first and come back to this.

Day 1Calibrate and fill the biggest gap

  1. Read how a staff engineer sounds — the four-beat answer20 min
  2. Take the quizzes on 10 terms you'd be embarrassed to get wrong (tokens, embeddings, attention, RAG, reranker, KV cache, eval gate, agent loop, prompt injection, cost per query)60 min
  3. Write down the 3 topics you scored worst on — they get extra time on days 2–410 min
  4. Coding: 3 AI-engineering problems (tokenizer, top-k/top-p sampling, cosine top-k)90 min

Day 2Foundations and serving maths

  1. LLM Foundations questions — answer each out loud before reading60 min
  2. Training, tuning & inference questions60 min
  3. Drill: KV-cache memory and concurrent users until you can do it without the calculator30 min
  4. Coding: attention with a causal mask, KV-cache decode loop60 min

Day 3RAG end to end

  1. RAG fundamentals and chunking questions60 min
  2. Retrieval & ranking, then production RAG90 min
  3. Coding: BM25 and Reciprocal Rank Fusion45 min

Day 4Agents and evals

  1. Agent fundamentals and agents in production90 min
  2. Evaluation and LLMOps90 min
  3. Coding: a minimal agent loop with a step budget; an eval harness with a paired bootstrap60 min

Day 5System design at full length

  1. Two full AI system designs, timed at 45 minutes each, using the outline on each page2 h
  2. Two estimation drills and one debugging drill60 min
  3. ML infrastructure questions you're weakest on45 min

Day 6Behavioural and your stories

  1. Read the STAR bank, then rebuild 6 stories from your own work: pushback, incident, trade-off, influence, failure owned, changed your mind2 h
  2. Behavioural questions — rehearse each story against the Mid/Senior/Staff ladder60 min
  3. One practical multi-stage build, timed75 min

Day 7Mock and rest

  1. Full mock: one design (45 min), one coding (45 min), two behavioural (20 min) — record yourself2 h
  2. Clear your review queue30 min
  3. Read lines to have ready, then stop studying by early evening15 min

One day before

  • Re-read your 6 STAR stories; each has a number in the result.
  • Say the four-beat answer out loud once: frame the decision, give the default, name the measurement, name the failure mode.
  • Do the KV-cache and cost-per-query estimates once from memory.
  • Prepare 3 questions for each interviewer about how the team evaluates and ships model changes.
  • Sleep. A rested 80% beats an exhausted 95%.

Only have one day?

  1. The four-beat answer (20 min) — it changes every answer you give.
  2. One system design question at full length, using its outline and rubric (60 min).
  3. The KV-cache and cost calculators until the maths is automatic (30 min).
  4. Three STAR stories from your own work, each with a number (60 min).
  5. Two AI coding problems (60 min), then stop.
Knowledge check

…

Picking questions…

Buy the handbook a coffee

It's free and stays free. If it helped you learn or land a role, a coffee keeps new chapters coming.

Secure checkout by Stripe — card, Apple Pay or Google Pay. No account needed.

Help improve the handbook

Spotted a mistake, missing topic or confusing page? Tell us — it goes straight to the editor.

Sent with this page: /interview/plan/