Intervue.ai
Voice AI interviewer with resume-aware prompting, speech recognition and live code.
What it is
A mock interview tool that talks back. You upload a resume and a job description, and it runs a voice interview that follows up on your actual answers instead of reading from a fixed list of questions.
How it works
GPT-4.1-mini drives the conversation, with the resume and job description in the prompt so the questions are specific to you. The loop is speech recognition, then silence detection to work out when you have finished talking, then the model, then text to speech back. That round trip lands at about 1.8s median, roughly 2s end to end, in four languages.
There is a Monaco editor in the same session so the coding round does not send you to another tab.
The part that took longest
Silence detection. If it cuts in too early it interrupts you mid sentence. If it waits too long the conversation feels dead. Getting that threshold right is most of the difference between something that feels like a person and something that feels like a form with a microphone.
Resume handling
Resumes get parsed out of the PDF and hashed with SHA-256, so uploading the same file again is recognised instead of reprocessed. That is what makes reuse across 100+ sessions cheap rather than linear in cost.
What I would fix
Most of the latency is the model round trip. Streaming partial responses straight into TTS instead of waiting for a complete answer would cut how slow it feels, even if the median number does not move.