AI Mock Interview

Most candidates practice interviews in a vacuum — no realistic pressure, no calibrated feedback, no signal on what a real interviewer actually values. Existing tools either give generic tips or require a human coach (expensive, hard to schedule).

Built an end-to-end interview simulation platform. The LLM plays a senior engineer from a specific company persona, asks adaptive follow-ups, and at the end scores the candidate's responses against a rubric (correctness, communication, depth, speed). Feedback is streamed back as structured markdown in under 10 seconds. Submitted and won at a 48-hour hackathon.

• 01

  Rubric-grounded evaluation with chain-of-thought reasoning produces far more actionable feedback than open-ended scoring.

• 02

  Streaming the feedback (not waiting for the full response) makes the UX feel dramatically faster — critical for a high-stakes moment.

• 03

  System prompt persona injection alone wasn't enough; few-shot examples of company-style questions were needed to stay in character.

• 04

  Time pressure in hackathons forces the right scope decisions — cut the nice-to-haves early.