Using AI to analyse job interviews
Today I’ll show you a great example of a game-changing use of AI: job interviews.
To make this work, you’ll need a transcript of the whole interview (a text version). There are different ways to get one. Some online meeting tools let you record meetings easily; in others you’ll need to add an AI agent to the meeting. Even a medium-quality transcript like the one from Google Meet will do (it struggles with several languages and IT terms in one conversation). But the better the quality, the less distortion you’ll get.
Then, with ChatGPT or your favourite AI assistant and a model with a large context window, we produce a detailed analysis of the interview:
- An overall summary and assessment of the candidate from different angles.
- An assessment of the interview style and of the interviewer’s work: what was great and what could be improved next time.
- An assessment of the tone of the candidate’s answers and how confident they were on each topic discussed.
This fully solves the problem of giving good feedback to the candidate and the recruiter, helps align interviewers across the organisation and develop their skills, and gives a more objective, less emotional assessment of the candidate.
Now I can formally describe my own interview style for senior positions:
Format. The interview is structured as a conversation in which the interviewer (Mikalai) asks both behavioural and deep technical questions.
Tone. The tone is generally informal and friendly, but with elements of a “stress” interview, which makes it easier to see how the candidate behaves under pressure or in an uncomfortable situation.
Delivery. Mikalai runs the interview in a “challenge, discussion, feedback” style: he constantly challenges the candidate’s statements, asks for clarification and draws them into practical examples.
And here are his strengths (of course, I won’t show you the weaknesses):
Deep technical dive. The interview covers the key topics for a senior level: performance tuning, threading models (Tomcat vs Netty), databases, caching, Spring Boot (starters, auto-configuration), service architecture, reactive vs blocking approaches.
Testing how the candidate thinks. The candidate isn’t just asked technical questions, but made to think, discuss, defend a position or admit the weak points of their arguments.
Revealing depth of knowledge. The interviewer isn’t satisfied with superficial answers and keeps digging for real understanding, which separates “memorised theory” from real experience.
A format close to real work. For example, the service case isn’t just a theoretical question but resembles a real task, which makes it possible to assess the candidate’s engineering approach.