// interview prep
AI engineering interview prep
Interviews test judgment, not trivia: can you design, debug, and reason about systems that use models? Here is the free hub to get ready — role map, topics, a question bank, and system-design drills.
AI Engineer
Builds apps on top of models. Interviews test RAG, agents, system design, and production judgment.
GenAI Developer
Ships LLM features. Expect prompts, structured outputs, function calling, and evals.
Agentic AI Engineer
Builds agents. Expect tool use, memory, orchestration, guardrails, and when NOT to use an agent.
Forward Deployed Engineer
Customer-facing delivery. AI engineering plus discovery, scoping, demos, and communication.
Chunking, embeddings, reranking, grounding, citations.
studyTools, memory, control loops, and workflow vs agent.
studyTest sets, error analysis, tracing, regression gates.
studyToken budgets, caching, model tiering, LLMOps.
studyInput validation, output constraints, prompt-injection defense.
studyAI Engineer vs ML Engineer vs GenAI Developer.
studyReal questions with the shape of a strong, developer-level answer. Try to answer before you expand.
Your RAG app returns confident but wrong answers. How do you debug it?
Start with retrieval, not the model. Log the retrieved context — was the right chunk even fetched? Then check chunking, reranking, and grounding. Only after that, adjust the prompt and model.
How would you evaluate a RAG system?
Build a small representative test set, do error analysis (read outputs, label failures, group them), add automated checks for the failure modes you find, and track a score over time to catch regressions.
When would you use an agent instead of a workflow?
Only when the steps can't be predetermined. Workflows are more reliable and cheaper to run; reach for a full agent for open-ended, branching, multi-tool tasks — and add guardrails, stopping conditions, and evals.
Design a chat-with-your-docs feature for 100k users.
Cover the layers: ingestion pipeline, structured chunking, vector store + metadata, retrieval + reranking, grounded answers with citations, caching, cost/latency budgets, evals, tracing, and prompt-injection defense. Draw the diagram and justify trade-offs.
Your token bill tripled overnight. What do you do?
Trace usage per request to find the driver, add semantic caching for repeated calls, trim retrieved context, and tier models (use smaller models where quality allows). Treat cost as an architecture decision, not an afterthought.
How do you defend a tool-using agent against prompt injection?
Treat retrieved/user content as untrusted, validate inputs, constrain outputs, give tools least privilege, require confirmation for high-impact actions, and separate instructions from data in the prompt. Test with red-team cases.
More in the full write-up: AI engineer interview questions.
- 01Design a chat-with-your-docs feature for 100k users.
- 02Design an agent that books meetings across calendar and email tools.
- 03Design an eval pipeline that gates every deploy.
- 04Design a cost-aware RAG serving layer with caching.
- 05Design observability for a multi-step agent.
For each: draw the layers, name the trade-offs, and say how you'd evaluate and control cost.
- Bring 3 real projects, each with a README, an architecture diagram, a short demo, and a note on trade-offs. See 5 AI projects that get you hired.
- Write resume bullets as outcome + system + metric(“cut answer latency 40% with semantic caching in a RAG service”).
- Rehearse the narrative: how to become an AI engineer walks the full path.
Want the full prep, structured?
The AI Engineer Interview & Portfolio Kit bundles an extended question bank, project specs, system-design frameworks, and resume templates into one focused kit.
Production AI Notes
One practical AI engineering email each week
One concept, one architecture, one project idea, and one interview question — written for developers who want to build and ship real AI systems.