MockView

Settings
Home Page
Sample Report
Progress Report
Live Interview

Inspiration

A friend told me he was practicing interview questions by asking generative AI to type out questions for him; I wondered if there was a more streamlined way to automate it, and even better, for it to seem responsive and give feedback.

What it does

MockView is an interactive interview prep tool that features real-time tracking of WPM and filler words, feedback on weak answers, and ElevenLabs integration to hear the ideal tone and pacing of your corrected response.

How we built it

We planned before we coded: the system architecture (speech-in, transcript, live metrics, report-out), the UI/UX across three screens (setup, live interview, report), and the data structures behind them. Then we built it with browser speech recognition for voice input and ElevenLabs for the interviewer’s voice and the ideal-delivery playback. We deployed on Vercel and used Claude as a development tool for code and visual layout.

Challenges we ran into

Real-time browser APIs were the hardest part. Browser speech recognition tends to stop after pauses, so keeping it listening continuously took trial and error. The interviewer’s voice, the mic, and the live metrics (WPM and filler counts calculated mid-sentence) also had to hand off smoothly. We added fallbacks, such as typed answers, when the mic is unavailable.

Accomplishments that we're proud of

We shipped a working app as a first-time hacking team and still had time to polish it. That includes three interviewer personalities, a live feedback panel, highlighted transcripts, and a sample report you can open without running a full interview.

What we learned

We learned what browser speech recognition can do out of the box and where it breaks down when it has to listen continuously and respond in real time. ElevenLabs showed us how much a natural voice turns a quiz into a conversation, and how useful it is to hear what a strong answer sounds like. We also learned to co-pilot with LLMs. Because we planned the architecture and data structures first, we could give Claude well-scoped tasks and check its output against our plan, using it effectively as a tool and not a project lead.