AI Engineer Interview Questions

---

🎯 Start here: The AI Engineer 75

The AI Engineer 75 is this repo's Blind 75: the 75 highest-signal items across everything here, in checkbox form.

What	How many	Bar to clear
Must-know questions	64	Answer out loud, cold, before opening the answer
System design mocks	4	45-60 minutes each on a doc before reading the solution
From-scratch coding challenges	7	Working implementation before peeking at the reference

If you can clear all 75, you are ready for most AI engineering loops. If you can't, the gaps tell you exactly which topic to study. Everything else in this repo is the depth behind this list.

How much prep do you actually need?

Your runway	Do this	Volume
Days	The AI Engineer 75, then CHEATSHEET.md the night before	75 items + one evening review
1 week	The 75 + the 1-week cram plan	75 items + 6 crash courses
2-4 weeks	The 4-week plan: all 13 crash courses, full question banks on your weak topics, 3 design mocks	200+ questions + 3 case studies + all 19 challenges
1-2 months+	The 8-week plan: all 541 topic questions, all 10 case studies, all 19 challenges, plus a portfolio project	Everything, plus building
Targeting a company	Add that company's interview questions from the banner above	+10-12 tailored questions and the loop map
Not an "AI Engineer" title	Start from your role guide instead	Role-calibrated study map

Essential shortcuts: The AI Engineer 75 · Glossary · Mock interview kit · Night-before cheat sheet · Study plans · Curated papers and courses · aidaddy.tech, the companion site for AI system design and interview prep

---

Company interview questions

Interview questions, loop maps, and prep priorities for 33 companies, tiered from frontier labs to applied AI shops and spanning the US, Europe, China, and India, including inference-silicon, vertical AI, autonomy and robotics. Everything is built from public information (job postings, engineering blogs, technical reports, published interview reports), not leaked material, and each page has a "last reviewed" note and sources.

Company	Tier
Anthropic	Frontier lab
OpenAI	Frontier lab
Google DeepMind	Frontier lab
Meta	Frontier lab
xAI	Frontier lab
Mistral AI	Frontier lab (Europe)
DeepSeek	Frontier lab (China)
Moonshot AI	Frontier lab (China)
Zhipu AI	Frontier lab (China)
Sarvam AI	Frontier lab (India)
Microsoft	Big tech
Amazon	Big tech
Apple	Big tech
NVIDIA	Big tech
Qwen (Alibaba)	Big tech (China)
Databricks	AI-native & infra
Scale AI	AI-native & infra
Perplexity	AI-native & infra
Cursor (Anysphere)	AI-native & infra
Cohere	AI-native & infra
Hugging Face	AI-native & infra
Together AI	AI-native & infra
Glean	AI-native & infra
Palantir	Applied / forward-deployed
Sierra	Applied / forward-deployed
Harvey	Applied / vertical AI (legal)
Abridge	Applied / vertical AI (health)
Waymo	Applied / autonomy
Figure AI	Applied / robotics
Cognition (Devin, Windsurf)	AI-native & infra
Groq	AI-native & infra
ElevenLabs	AI-native & infra
Character.AI	AI-native & infra

Role guides

AI questions now show up in backend, frontend, product, data, DevOps, QA, mobile, and security loops, and the depth expected varies a lot by role. Each guide maps how that role's interviews changed, what you are actually expected to know (and what you are not), and 12-14 role-specific questions.

Role	Guide
Backend Engineer	Guide
Frontend Engineer	Guide
Product / Full-stack Engineer	Guide
Forward Deployed Engineer	Guide
Data Engineer	Guide
DevOps / Platform / MLOps Engineer	Guide
QA / SDET Engineer	Guide
Mobile Engineer	Guide
Security Engineer	Guide
"Am I an ML Engineer or AI Engineer?"	Guide

Topics

Each topic has a crash-course primer (README.md) and a full question bank with worked, collapsible answers (questions.md). Work with the answer collapsed until you have tried it yourself.

#	Topic	Questions	What it covers
01	ML & Deep Learning Foundations	50	Bias-variance, optimization, regularization, metrics, loss functions, the fundamentals a fine-tuning or evals answer is built on
02	LLM & Transformer Fundamentals	60	Attention, positional encodings, tokenization, scaling laws, MoE, decoding, KV cache, reasoning models
03	Prompt Engineering & Context Engineering	45	Few-shot design, chain-of-thought, structured outputs, prompt caching, context rot and compaction
04	RAG & Retrieval	55	Chunking, embeddings, hybrid search, reranking, agentic RAG, retrieval evaluation
05	Fine-tuning, RLHF & Alignment	51	SFT, LoRA/QLoRA, DPO/PPO/GRPO, distillation, GPU memory maths for training
06	Agents, Tool Use & MCP	60	Tool calling, MCP, planning patterns, multi-agent design, agent evaluation and security
07	Evals & Observability	50	LLM-as-judge, benchmark limits, RAG and agent evals, tracing, regression testing
08	Inference, Serving & Production LLM Systems	51	Prefill vs. decode, KV cache paging, quantization, speculative decoding, cost engineering
09	Safety, Security & Responsible AI	47	Prompt injection, OWASP LLM Top 10, guardrails, agent security, data governance
10	Multimodal Models	35	Vision-language architecture, diffusion, ASR/TTS, voice agents, multimodal RAG
11	AI System Design	10 case studies	A reusable answer framework plus eight worked case studies
12	Coding Challenges	19 challenges	Implement attention, BPE, sampling, KV cache, an agent loop, and more, from scratch
13	Interview Process & Behavioral	37	Loop anatomy by company type, take-homes, portfolio projects, AI-specific behavioural questions

System design case studies

Full worked examples in 11-ai-system-design/case-studies, each following the same template: requirements, architecture, evaluation plan, cost estimate, failure modes, and likely follow-ups.

#	Case study
1	Enterprise RAG Assistant
2	AI Code Assistant
3	Customer Support Agent
4	Semantic Search at Scale
5	Content Moderation Pipeline
6	Document Intelligence Pipeline
7	Text-to-SQL Agent
8	Meeting Assistant
9	Real-time Voice Agent
10	LLM Gateway and Serving Platform

Coding challenges

Nineteen self-contained Python files in 12-coding-challenges, numpy and the standard library only, each with a reference solution and a real test suite. Read only the problem statement, implement it yourself, then run the file directly, for example python3 12-coding-challenges/01_attention.py.

#	Challenge	Difficulty	Concepts
01	Attention	Medium	Scaled dot-product attention, multi-head, causal masking
02	BPE Tokenizer	Medium	Byte-level BPE: train, encode, decode
03	Sampling Strategies	Easy	Temperature, top-k, top-p, min-p, repetition penalty
04	Positional Encodings	Medium	Sinusoidal PE, RoPE
05	LayerNorm & Softmax	Easy	Stable softmax/log-softmax, LayerNorm, RMSNorm
06	KV Cache	Medium	Autoregressive decode loop with vs. without KV cache
07	Mini-GPT Forward Pass	Hard	Full GPT block forward pass: embeddings, attention, MLP, logits
08	Semantic Search / RAG	Medium	Embed, index, cosine retrieval, context assembly
09	Text Chunking	Easy	Fixed, sliding-window, recursive, sentence chunkers
10	Agent Loop	Medium	Tool-calling agent loop with a mock LLM
11	Rate Limiter & Retry	Medium	Token bucket, exponential backoff with jitter
12	Eval Metrics	Medium	pass@k unbiased estimator, QA F1/EM, judge harness skeleton
13	Streaming Parser	Hard	SSE parser, incremental tool-call argument assembly
14	LoRA Adapter	Medium	LoRA forward, merge, batched multi-adapter serving
15	Beam Search	Medium	Beam search with length normalisation
16	Semantic Cache	Medium	Exact and semantic tiers, TTL, LRU, tag invalidation
17	Hybrid Search & Rerank	Medium	BM25, reciprocal rank fusion, MMR
18	Constrained JSON Decoding	Hard	Schema-masked token decoding
19	Speculative Decoding	Hard	Draft-and-verify rejection sampling

Contributing

Corrections, new questions, and new case studies are welcome. See CONTRIBUTING.md for the format.

---

License

MIT.