HN Showcase
Back to browse
I built an AI-powered technical interview prep tool

I built an AI-powered technical interview prep tool

by wa5ina·Mar 9, 2026·1 point·1 comment
Visit ProjectView on HN

AI Analysis

●MidShip ItSolve My Problem

WebRTC voice interviews with Claude scoring, but Pramp exists.

Strengths
  • •Sub-100ms WebRTC voice latency creates actual conversation flow, not turn-based chat
  • •Monaco Editor means you practice in the same environment you'll code in daily
  • •Five-dimension scorecard pinpoints exact moments where you lost points
Weaknesses
  • •AI interview prep is saturated — Pramp, Interviewing.io, HelloInterview all compete here
  • •Author admits AI still gets pedantic about syntax over algorithmic logic
Category
Education
Target Audience

Software engineers preparing for technical interviews

Similar To

Pramp · Interviewing.io · HelloInterview

Post Description

LeetCode wasn't actually preparing me or anyone else for the pressure of a real technical interview.

Most platforms just give you a text box and a compiler. I wanted something that actually simulates the back-and-forth of an interviewer pushing you on time complexity, asking follow-up questions, and pointing out edge cases you missed.

> The Stack: I built the backend with nestjs, prisma, postgresql, Webrtc for real time back and forth conv and [socket.io](http://socket.io) events in panel like code run etc…. claude-haiku-4-5 for conv and anthropic.claude-sonnet-4-6 for interview scoring.

> Where it still falls short: It’s not perfect. Sometimes the AI gets overly pedantic about syntax instead of focusing on the core algorithmic logic. I’m still tuning the prompts to make it behave more like a senior engineer and less like a strict compiler.

I'm currently trying to stress-test the system. I’d love for you to try breaking the AI interviewer or tell me why this approach is flawed. Happy to answer any questions about the stack or the prompt engineering.

Similar Projects