Check out our technical paper here: https://arxiv.org/abs/2501.15627 and the video demo: https://www.youtube.com/watch?v=Op8hyLW7Z84

I’ve been obsessed with the art of the interview since I was in college. In my career I interviewed over 100 people and was interviewed from tech companies from startups to big tech and hedge funds.

I built Neuraprep because I noticed something missing — while software engineers have leetcode.com and finance folks have quantquestions.com, other engineering domains (like ML, data science, MLOps) don’t have a go-to platform to prep for interviews. Sure, there’s Kaggle and Coursera, but nothing unified.

So I spent a summer collecting 400+ real interview questions and built detailed answers for each. I drew from academic sources, online communities, and LLMs to refine the content. Then I used this dataset to build an AI that mimics how a human interviewer evaluates responses.

Here’s how it works:

• The reasoning engine extracts core ideas from the user’s answer. • It compares them to the expected ideas from the database. • If something is missing, the conversation continues — just like a real technical interviewer would do.

With recent voice and reasoning model advancements (thanks Sesame, O3), it now runs on-demand phone interviews that feel surprisingly real.