Interview preparation
Prepare for interviews honestly
An interview is not an exam with a list of right answers. It is a conversation between people who want to find out if they can work well together. This page is our advice, from interviews on both sides of the table. The role guides then show you how real technical interviews go, turn by turn.
1
An interview is a conversation in both directions
The company wants to learn who you are and how you work. You also want to learn about them: the work, the team, the people you will report to, and how they make decisions. Both sides decide. If you think of the interview this way, it feels less like a trial, and you listen better.
So prepare your own questions, and ask them. Good questions are about the work. For example: "What would I work on in my first three months?", "How do you know that an AI feature works well before you release it?", "Who is on call when it fails at night?" and "What does a good first year look like in this role?"
Listen to the answers carefully. A clear, honest answer is a good sign. A vague answer is also information. You are choosing where you spend most of your waking hours.
2
Do not use AI or other help to cheat in a live interview
We have seen this first-hand, and we have read many stories from other interviewers: candidates who read answers from an AI tool during a live interview, or who get help from another person who listens off screen.
We understand the pressure. Months without a job are hard. When a family depends on you, every interview feels like the last chance, and a shortcut can look reasonable. We do not judge the pressure. But we ask you not to do this, because it destroys the chances that you really have.
Interviewers notice more than you think. A pause before every answer. Eyes that move to another part of the screen. Long, perfect, general answers that fail at the first follow-up question, such as "Why did you choose that?" about your own project. A tool can give you words. It cannot give you your own experience, and good interviews are built to find that experience.
When an interviewer suspects help, they stop believing everything else, including the parts that were true and good. Companies now say clearly that they disqualify candidates who use AI tools when it is not allowed, and many have brought back interviews in person. And if it works, you start a job that you cannot yet do, with more pressure than before. A reputation for honesty, on the other hand, follows you between teams and helps you for years.
What to do instead: say what you know, and say clearly what you do not know. "I have not used that in production. This is how I would start, and this is what I would check" is a good answer. Interviewers trust it more than a perfect answer. Some companies now allow AI tools in some rounds, such as a take-home task or a pair-programming session. That is fine: follow the rules of each round, use the tool openly, and if you are not sure what is allowed, ask.
Use AI as much as you like to prepare: ask it to play a strict interviewer, to question your project, or to find the weak lines in your CV. That is learning, not cheating. And if you need a job soon, apply widely, include roles one step below your target, and ask recruiters honestly about their timeline. A real chance is worth more than a borrowed answer.
See how this looks from the interviewer's side in The polished CV, one of the interviews in our AI Engineer guide.
3
Build real skills, and learn to explain them
Every line on your CV is an invitation to a question. If you write "built a RAG system", expect to explain how you split the documents, why you chose that model, how you measured the answers, what failed, and what you would change. A short CV that you can explain in depth is stronger than a long one that you cannot.
Practise saying your work out loud: the problem, the options you considered, why you chose one, the result in numbers, and the limits. Our lesson Explain decisions shows how, andThe same person, told honestlyshows how a small, true CV can do better than a large one.
4
Your English does not need to be perfect
Many people in technical interviews speak English as their second or third language. That includes many interviewers. A good interviewer listens for your thinking, not for perfect grammar.
So speak simply. Short sentences are fine. If you do not understand a question, ask: "Sorry, can you say that again?" or "Do you mean the training data or the live data?" If you lose your point, say so, and start again. If you need a moment to find a word, take it. All of this is normal, and it is much better than a long answer to the wrong question.
Practise out loud before the interview, in English, about your own projects. The words that you need come faster when you have said them a few times. Our role guides include candidates who are not native speakers, so you can see how this looks in a real conversation.
5
No work experience in the role? Show work that is close to real
You do not need a job to show that you can do a job. You need one or two projects that have the difficulty of real work: messy data, a measurement of quality, failure cases that you found and handled, a deployment that someone else can use, and a README that explains your decisions and limits.
A chatbot from a tutorial does not show much, because thousands of people have the same one. A system that you measured, broke, and improved shows a lot, because the interview can go deep into it and you have answers. See Portfolio anatomy for what a strong project contains, and The self-taught builder for an interview about a project like this.
6
There is no formula that passes every interview
Interviews are different in each company, each team and with each interviewer. Be careful with anyone who says "Learn these 50 questions and you will pass 100% of interviews". Nobody can promise that, because nobody knows who will interview you and what their team needs.
The same is true for the noise about AI tools. Almost every week, someone says that a technology is dead: "RAG is dead", "prompt engineering is dead", "fine-tuning is dead". Usually the tool changes, and the problem that it solves stays. Learn the problem, the trade-offs and how to measure the result. Then you can discuss any new tool with confidence, including in an interview.
7
Interviewers are not always right. You can disagree, with reasons
Interviewers are people. Sometimes a question has a wrong assumption in it, or the interviewer understands a concept only in part. We have seen this in real interviews.
Their position is different from yours: they decide, and they often know things about the team that you do not. So be respectful. But you do not have to agree with something that you know is wrong. Say how you see it, give your reason or an example, and stay open. For example, when an interviewer says that long inputs make search unnecessary: "I see it a bit differently. A long input works, but at 9,000 documents a day it costs a lot, and some facts get missed. Could we test both on 50 documents and compare?"
A good interviewer values this, because it is how you would work in their team. If an interviewer punishes a polite, reasoned disagreement, that also tells you something about the team.
See how Daniel disagrees with a hiring manager, and how Leila corrects an interviewer about her own field.
8
Luck is part of it. Do not give up, and build your network
There is always some luck in hiring: meeting the right people at the right time. A rejection is often not a judgement of you. Another candidate had the exact domain experience, the role was filled from inside the company, or the budget changed. Do not let one rejection, or ten, tell you who you are.
You can make your own luck larger. Share what you build. Write short posts about what you learned and what failed. Go to meetups, and help people in communities. Ask people for advice, not for a job. When someone knows your work, your application is read in a different way. And a "no" for one role can lead to another role later, with the same people.
After each interview, write down the questions that were hard while you still remember them. Ask politely for feedback (not every company gives it). Then practise those questions.
9
Keep learning, but stay focused
Choose a target role. Read ten job descriptions for it, and write down what repeats. Learn those things deeply, and build one project that shows them. Do not try to learn every new tool that is announced. Depth in what the role needs beats a long list of tools that you tried once.
Our learning paths and the course AI Career Foundations help you choose a role and plan the work.
How serious interviews really go
Most good technical interviews start by listening. "Tell me about yourself." "Walk me through a project that you are proud of." Then the interviewer goes deeper into what you said, one level at a time: what you did, how, why, what else you considered, what went wrong, and what you would do at ten times the size. They continue until they find the edge of what you know.
It is normal to be nervous, to ask for a moment to think, and to think in silence. Interviewers expect it. Reaching the edge is not a failure either. Everyone has one. The interviewer learns the most from how you behave there: do you say "I do not know", do you reason from what you do know, and do you stay honest? Our role guides are built around this. You read a real-world job description, and then you follow different candidates through the same interview, from the first answer to the edge.