Skip to contentSkip to footer

Ai Software Engineer Interview Questions

Salaries
Interviews

5,953 ai software engineer interview questions shared by candidates

1) The first round was an hour long interview which consisted of a introduction and talk about your background and internship/project experience relevant to the role (for me this was an RL course project). Then this moved into specific RL questions (e.g. describe DQN, PPO, AC, which algorithm to use for discrete/continuous action spaces, how can DQN be adapted for a continuous action space). Next, they asked to formulate an RL setup specific to the team you are interviewing with like articulating the state&action space, the reward, the agent algorithm selection (definitely prepare for that). Expect follow up questions. Then it moved into ML questions (bias-variance trade-off, regularisation strategies, vanishing gradients and how to deal with them, do you know what JAX is, what is JIT compilation, questions were general ML). Then there were some general python questions (slicing, what is a decorator). Then some data structures questions (loops in a linked list, cycles in a graph). Get ready to be grilled. 2) Live coding: Implement 3 functions for a Tabular Q-Learning algorithm in a grid world (step, select_action, update). 4x4 grid world, agent starts at (0,0) and needs to get to (3, 3), zero-indexed. Step takes in a state and action and returns the new state. The main thing is to handle invalid actions like stepping off the grid. select_action is standard eps-greedy. update is standard Q-Learning update. Then there are some analysis questions after you implement the algorithm which I didn't have enough time to get to (probably why I didn't get it). After several emails asking for an update after the second interview, they replied saying it had been filled. My biggest recommendation is not to waste too much time preparing for overly complex possible questions. Should have basic RL coding at your finger tips but be able to talk about more complex concepts at a high level. Was good RL & ML review for me but I'm pretty disappointed about spending so much time preparing, seemingly doing ok and getting nothing back. Cool company overall.
avatar

AI Research Intern

Interviewed at InstaDeep

3.7★
Jan 24, 2026

1) The first round was an hour long interview which consisted of a introduction and talk about your background and internship/project experience relevant to the role (for me this was an RL course project). Then this moved into specific RL questions (e.g. describe DQN, PPO, AC, which algorithm to use for discrete/continuous action spaces, how can DQN be adapted for a continuous action space). Next, they asked to formulate an RL setup specific to the team you are interviewing with like articulating the state&action space, the reward, the agent algorithm selection (definitely prepare for that). Expect follow up questions. Then it moved into ML questions (bias-variance trade-off, regularisation strategies, vanishing gradients and how to deal with them, do you know what JAX is, what is JIT compilation, questions were general ML). Then there were some general python questions (slicing, what is a decorator). Then some data structures questions (loops in a linked list, cycles in a graph). Get ready to be grilled. 2) Live coding: Implement 3 functions for a Tabular Q-Learning algorithm in a grid world (step, select_action, update). 4x4 grid world, agent starts at (0,0) and needs to get to (3, 3), zero-indexed. Step takes in a state and action and returns the new state. The main thing is to handle invalid actions like stepping off the grid. select_action is standard eps-greedy. update is standard Q-Learning update. Then there are some analysis questions after you implement the algorithm which I didn't have enough time to get to (probably why I didn't get it). After several emails asking for an update after the second interview, they replied saying it had been filled. My biggest recommendation is not to waste too much time preparing for overly complex possible questions. Should have basic RL coding at your finger tips but be able to talk about more complex concepts at a high level. Was good RL & ML review for me but I'm pretty disappointed about spending so much time preparing, seemingly doing ok and getting nothing back. Cool company overall.

Viewing 5841 - 5850 interview questions

Glassdoor has 5,953 interview questions and reports from Ai software engineer interviews. Prepare for your interview. Get hired. Love your job.