Amazon

English · Русский · AI Interview Atlas › Companies

Big Tech · International · Loop reviewed 2026-10-05

Amazon’s official preparation page describes a general TPM process. The AI-product path remains a secondary guide, and Engineering topics come from a compilation. The general TPM stages do not establish an AI-specific loop.

On this page: Interview loop · Questions (12)

Roles covered

• AI Engineering: AI / LLM engineer
• AI Leadership: AI product manager (senior to group), Technical program manager

Interview loop

Stage	What happens	Basis
General TPM technical phone screen Technical program manager	The official TPM preparation page describes a 60-minute screen with behavioural, situational and technical discussion; a second phone screen may follow. This is general TPM guidance, not an AI-specialist process.	✅ confirmed by the company TPM Interview Prep, Amazon Jobs, retrieved 2026-10-05
General TPM writing assessment and interview loop Technical program manager	The page describes a writing assessment before a loop of five 55-minute interviews, including at least one systems-design question. Management preparation covers metrics, dependencies, escalation and tradeoffs. These published stages describe the general TPM path at retrieval.	✅ confirmed by the company TPM Interview Prep, Amazon Jobs, retrieved 2026-10-05
AI product manager path	A recruiter screen on ownership and customer obsession, a one- to two-page written answer to a Leadership Principles prompt, product screening rounds, then a final loop of four or five rounds with Leadership Principles deep dives, product thinking, generative-AI evaluation, system design for technical PM roles and a Bar Raiser.	† prep guide or compilation Amazon AI Product Manager Interview Guide, Aced (formerly Exponent), retrieved 2026-09-26
Engineering questions	A compilation attributes to Amazon questions on implementing attention, serving cost, parallelism, enterprise RAG design and diagnosing latency regressions.	† prep guide or compilation AI Engineering Interview Questions Company Wise, Outcome School, published 2026-09-19, retrieved 2026-09-26

Coding: unknown. The retained secondary sources do not meet the atlas's evidence threshold for a coding requirement. (Amazon AI Product Manager Interview Guide, Aced (formerly Exponent), retrieved 2026-09-26, AI Engineering Interview Questions Company Wise, Outcome School, published 2026-09-19, retrieved 2026-09-26)

Questions reported for Amazon

LLM fundamentals

• Derive the memory required by a decoder's KV cache as batch size and context length grow.
  • Knowledge · Asked at: Amazon †, OpenAI † · ✍ Answer
  • Tests: Whether cache size is calculated from the actual architecture.
  • A strong answer covers:
    • For equal sequence lengths, use 2 × batch × layers × tokens × KV heads × head width × bytes.
    • Explain why queries are not retained for future decode steps.
    • Account separately for weights, activations, padding and allocation overhead.
  • Read: Efficient Memory Management for Large Language Model Serving with PagedAttention (Kwon et al., arXiv) · GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints (Ainslie et al., arXiv)

Inference, serving and cost

• Compare tensor, pipeline, data, sequence and expert parallelism for a large model deployment.
  • Knowledge · Asked at: Amazon †, Google and Google DeepMind †, Meta †
  • Tests: Whether communication patterns and memory limits determine the choice.
  • A strong answer covers:
    • Name each split and its traffic: TP per-layer all-reduce, PP stage activations, SP key-value exchange, EP token all-to-all.
    • Fit weights and KV first: TP inside the NVLink domain, PP across nodes, then traffic-free data-parallel replicas for throughput.
    • Measure per-GPU memory headroom, communication share of step time, pipeline bubbles and expert load imbalance at target latency.
  • Read: vLLM documentation (vLLM project) · Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity (Fedus et al., arXiv) · Efficiently Scaling Transformer Inference (Pope et al., arXiv) · Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism (Shoeybi et al., arXiv)
• You are asked to reduce serving cost by an order of magnitude. Rank the levers and explain how you would test whether that target is achievable.
  • Applied scenario · Asked at: Amazon †, Microsoft † · ✍ Answer
  • Tests: Whether savings are measured under explicit quality and latency constraints.
  • A strong answer covers:
    • Measure cost per successful task at the required quality and SLO.
    • Compare smaller models, shorter contexts, reuse, batching and precision changes.
    • Benchmark each change and check whether the combined savings actually reach the target.
  • Read: Optimizing inference (Hugging Face Transformers docs) · Site Reliability Engineering (Google)
• After a deployment, p99 latency doubles although model weights are unchanged. How do you isolate the cause?
  • Applied scenario · Asked at: Amazon †, Databricks †, OpenAI †, Perplexity † · ✍ Answer
  • Tests: Whether diagnosis separates queueing, scheduling, dependencies and traffic changes.
  • A strong answer covers:
    • Compare traffic mix, queue time and per-stage latency before and after release.
    • Inspect cache hit rates, batching, dependency tails and resource saturation.
    • Reproduce one hypothesis at a time and use a rollback or canary to verify causality.
  • Read: The Tail at Scale (Dean and Barroso, Google Research)

AI system design

• Design a knowledge assistant over ten million enterprise documents with per-user permissions and a continuously changing corpus.
  • System design · Asked at: Amazon †, Databricks †, Microsoft †, OpenAI † · ✍ Answer
  • Tests: Whether ingestion, retrieval, access control and serving form an operable system.
  • A strong answer covers:
    • Clarify query volume, freshness and permission semantics before sizing.
    • Separate ingestion and index versions from query-time authorisation and retrieval.
    • Evaluate grounded answers and test revocation, stale data and partial outages.
  • Read: Document-level access control (Microsoft Learn) · Contextual Retrieval in AI Systems (Anthropic)

Practical coding

• Implement scaled dot-product attention with a causal mask and tests that would expose attention to future tokens.
  • Coding · Asked at: Amazon †, Anthropic †, Google and Google DeepMind † · ✍ Answer
  • Tests: Whether shapes, masking and numerical stability are correct in executable code.
  • A strong answer covers:
    • Check batch, sequence and head dimensions before implementing score multiplication.
    • Use stable softmax after masking disallowed positions.
    • Test small hand-computable cases and ensure changing future inputs cannot change earlier outputs.
  • Read: Attention Is All You Need (Vaswani et al., arXiv)

AI product strategy and metrics

• Choose an AI agent or product you use and propose its most valuable improvement.
  • Product case · Senior · Asked at: Amazon † · ✍ Answer
  • Tests: Whether product judgement rests on observed friction and a testable hypothesis.
  • A strong answer covers:
    • Describe a real workflow and where the product fails the user.
    • Explain why the proposed improvement outranks alternatives.
    • Specify a prototype, success measure and possible adverse effect.
  • Read: Measuring the User Experience on a Large Scale: User-Centered Metrics for Web Applications (Rodden, Hutchinson and Fu, Google Research)

Programmes and delivery

• Three weeks before launching an LLM feature, evaluation reveals substantial hallucinations on edge cases. What happens next?
  • Applied scenario · Senior · Asked at: Amazon † · ✍ Answer
  • Tests: Whether you make a risk-based launch decision rather than protecting the date at any cost.
  • A strong answer covers:
    • Check severity, prevalence, affected users and the reliability of the evaluation.
    • Compare narrowing scope, adding review, fixing causes and delaying release.
    • Assign a decision owner, explicit exit criteria and a communication plan.
  • Read: AI Risk Management Framework (NIST) · Demystifying evals for AI agents (Anthropic)
• How would you roll out updates to Alexa devices already deployed in people's homes?
  • System design · Senior · Asked at: Amazon † · ✍ Answer
  • Tests: Whether you account for fleet diversity, interrupted updates and safe rollout.
  • A strong answer covers:
    • Clarify update scope, compatibility, connectivity and user disruption.
    • Use controlled cohorts, health checks and recoverable installation.
    • Define rollout pauses, rollback ownership and support communication.
  • Read: Site Reliability Engineering (Google)

Behavioral and values

• Tell me about a time you disagreed with someone and could not persuade them.
  • Behavioral · Senior · Asked at: Amazon † · ✍ Answer
  • Tests: Whether you work constructively after losing an argument.
  • A strong answer covers:
    • State both positions fairly and explain your evidence.
    • Describe the decision process and how you supported the agreed action.
    • Explain the result, limits of commitment and what you learned.
  • Read: Leadership Principles (Amazon Jobs)
• Tell me about an opportunity you saw that was larger than the original scope of your assignment.
  • Behavioral · Senior · Asked at: Amazon †
  • Tests: Whether you recognise broader value and justify taking on additional scope.
  • A strong answer covers:
    • Pick an opportunity you noticed from inside the assigned work, and explain what made it visible to you.
    • Show how you sized the extra value against the cost, and who agreed before you expanded scope.
    • Close with whether the original assignment still landed and what the larger value turned out to be.
  • Read: Leadership Principles (Amazon Jobs)
• Tell me about an AI decision you owned that had significant downstream consequences.
  • Behavioral · Senior · Asked at: Amazon †
  • Tests: Whether you consider impact beyond your immediate deliverable.
  • A strong answer covers:
    • Choose a decision whose effects reached people outside your team: users, operators, downstream systems or partners.
    • Show which second-order effects you anticipated, which you missed, and how you weighed them at the time.
    • Close with how you monitored the consequences after launch and what you changed in response.
  • Read: AI Risk Management Framework (NIST)

← OpenAI · Canva →