AI interview practice
Practise AI fundamentals, design trade-offs, and honest examples of AI-assisted work.
AI questions vary by company and role. Public guidance does not show that every general SDE interview includes a language-model round. Treat AI literacy as an added skill, and follow the role description and recruiter instructions.
Questions worth practising
| Prompt | What a clear answer covers |
|---|---|
| How does a language model produce text? | Tokens, context, likely next-token generation, and why an answer can be wrong |
| How would you answer questions from a changing document set? | Retrieval, permissions, source freshness, citations, and an honest no-answer path |
| How does model tool calling work? | The model proposes a call, application code validates and executes it, then returns the result |
| How would you test an AI feature? | Representative examples, expected behavior, failure cases, quality, latency, and cost |
| How do you use AI while coding? | One real task, bounded assistance, your review steps, and what you changed |
| Would you use AI in this interview? | Follow the interview instructions and ask before using an unapproved tool |
System design practice
Use one prompt and design it out loud. State the user need and constraints first. Then describe the request path, stored data, failure behavior, privacy controls, and measurements.
- Design a support assistant that answers from a changing help center.
- Add an AI summary to a service that processes uploaded documents.
- Build a safe tool that lets an assistant look up order status.
- Investigate a model feature whose latency or cost suddenly increases.
You do not need to build four projects. Use a small extension to one selected backend project only after its core behavior is complete. For example, add an assistant that searches project docs and summarizes synthetic incident logs. Include source links, a small evaluation set, permission checks, and a fallback when evidence is weak.
A four-step study order
- Read LLM basics and answer the questions without notes.
- Draw the request path in Build AI features.
- Practise the workflow example in AI-assisted workflow.
- Do a mock interview and explain one trade-off clearly.
Use the Google ML Crash Course for core concepts. Use the OpenAI function-calling guide as one provider’s implementation example. Review the OWASP Top 10 for LLM applications for current security risks.
Refresh before an interview
Read the job description again and identify whether the role is general backend, AI product engineering, ML infrastructure, or research engineering. Go deeper only where the role asks for it. Check current official docs because API names, model capabilities, and safety controls change over time.