interview-prep.io
AI mock interviews with secret feedback & model answers.
The Problem We Found:
Candidates memorized interview answers on paper, but froze or rambled for 3 minutes without stating clear metrics when speaking out loud.
How We Solved It in Code:
Built Web Speech API real-time transcript streaming paired with GPT-4o mini STAR color-coding. Spoken text is highlighted live: Situation (15%), Task (10%), Action (50%), Result (25%).
Technical Stack:
- Web Speech API + Tailwind HSL tokens + JSON-schema prompt parsing.
- Result: Reduces answer rambling by 40% in candidate practice sessions.
Check out the live build: https://interview-prep.io
#buildinpublic
Hey makers,
There is a wave of AI interview prep tools hitting the market right now. But as someone building one, I’ve had to ask myself: Can an AI really prepare someone for the subtle human dynamics of an interview?
After testing dozens of tools and building Interview Prep.io, here is my conclusion:
AI cannot replace human chemistry, but it is incredibly good at fixing structural habits you don't realize you have.
When people practice out loud with our simulator, we analyze their response structure. The most common issues we see are:
No "Result" in behavioral questions: People explain the problem (Situation) and what they did (Action), but completely forget the outcome.
Rambling: Taking 3 minutes to answer a question that should take 90 seconds.
We built a STAR-method visualizer that literally color-codes transcripts so users can see their structural gaps, along with a "Likelihood of Hiring" score.
I'd love to hear your thoughts on this:
Have you ever used an AI tool to prep for a real interview?
What did you feel was missing compared to doing a mock interview with a human friend?
If you'd like to test our tool, I'd love to get your feedback on our STAR scorecard.
-
2
The interesting opportunity isn't making AI behave more like an interviewer—it's helping candidates recognize the patterns that quietly undermine otherwise good answers. I'd keep validating whether users value the interview simulation itself or the structured feedback that changes how they answer every future interview. That's a much more defensible outcome.
Comment