What OpenAI, Anthropic, and DeepSeek actually ask AI engineers

What it does

This repository is a living study guide that collects publicly reported AI engineering interview questions from roughly 35 companies and organizes them by both topic and employer. It covers roles ranging from LLM inference engineer to MLOps, with each question tagged to the companies known to ask it and linked to explanatory answers wherever available.

The interesting bit

Instead of generic prep, it functions like an intelligence brief on the AI hiring market: you can see that DeepSeek and Moonshot AI ask about Multi-head Latent Attention, while Meta probes SwiGLU and pre-training tradeoffs. That specificity lets candidates study the exact failure modes and optimizations a given shop cares about.

Key highlights

• Covers frontier labs (OpenAI, Anthropic, DeepMind, xAI), big tech (NVIDIA, Apple, Amazon), infrastructure players (Groq, Together AI), and AI-native startups (Cursor, ElevenLabs)
• Topic breadth spans scaled dot-product attention, KV-cache memory math, continuous batching, speculative decoding, quantization, RAG, agents, and AI system design
• Many questions link to detailed companion blog posts on the underlying mechanics
• Explicitly sourced from public interview experiences; the README warns that loops change constantly and vary by team
• Maintained by Outcome School as a companion to a separate topic-wise question repo

Caveats

• Answers are still being added; the maintainers note that not every question has a linked explanation yet
• The README explicitly warns that interview loops change constantly and vary by team, level, and region, so this is a map of interests rather than a fixed script

Verdict

Bookmark this if you are interviewing for an LLM, AI platform, or research engineering role and want to reverse-engineer what each company values. Look elsewhere if you need hands-on projects or a guaranteed question bank.