AI Product Case Study
Practice out loud, get real feedback.
A real-time interview practice tool: GPT-4 asks and follows up, the browser transcribes your spoken answer live, and a performance dashboard tracks how each session went.
Live app
Next.js · GPT-4 · Web Speech API
2025
live speech transcript
GPT-4 follow-up
01 — The problem
What static prep tools miss.
✕Reading isn't rehearsing
✕No one probes deeper
✕No record to review
Format
A list of questions doesn't force you to actually speak an answer or think on your feet the way a real interview does.
Follow-up
Static question banks don't ask a natural follow-up based on what you just said, which is where real interviews get harder.
Feedback
Without a transcript or performance log, there's nothing to look back on after the session to see what to improve.
02 — The approach
How a session runs.
01 Questions
GPT-4 generation
Asks interview questions and follows up based on the substance of the previous answer.
02 Speech
Live transcription
The browser's speech recognition transcribes the spoken answer live.
03 Tracking
Performance log
An analytics dashboard logs each session for a record across practice runs.
03 — Under the hood
Two engineering decisions, up close.
Frontend
Next.js, deployed on Vercel.
The app ships as a Next.js application, deployed directly on Vercel so the live link always reflects the latest build.
framework: Next.js
deploy: Vercel, on every push
Voice
Browser-native speech, no extra service.
Speech-to-text runs through the Web Speech API in the browser, keeping the answer flow real-time without a separate transcription backend.
SpeechRecognition.start()
onresult → live transcript
04 — Honest scope
What this app doesn't do yet.
Browser-dependent speech
Transcription quality depends on the browser's built-in speech recognition, which varies across browsers and accents.
No persistent accounts
Session history lives in the current session rather than a signed-in, cross-device account.
Accounts
Add sign-in so practice history persists across devices and sessions.
Scoring
Layer structured feedback on top of the transcript — clarity, pacing, and completeness.