Inspiration
The idea came from my teammate Jacob. We started talking about how much interview skills matter, for college admissions, a first job, or really any moment where someone has to sell themselves in a room, and realized almost nobody gets real practice at it.
The numbers backed up what we already sensed. Ninety three percent of candidates report interview related anxiety, and freezing up under a tough question, not lack of qualification, is one of the most common fears people carry into an interview. Candidates typically spend five to ten hours preparing for a single interview, yet only about twenty four percent walk away feeling good about how it actually went. That gap between effort and outcome is what pulled us in.
Digging further, we found the real blind spot. Most prep time goes into rehearsing what to say. Almost none goes into understanding how it sounds, tone, pacing, filler words, the things that actually shape how confident someone comes across. Mock interviews with a mentor or counselor are hard to schedule, and not everyone has access to that kind of feedback in the first place. Most people end up practicing alone in front of a mirror, or not practicing at all, and they never find out they were rambling or leaning on "um" until the interview that actually counted.
That is what led us to build VoxCoach AI.
What it does
VoxCoach AI is a voice powered interview practice platform for college admissions and job interviews. A user picks a question set, records an audio answer, and the platform analyzes vocal tone, pacing, and filler word frequency from that recording. That analysis turns into a personalized feedback report with specific strengths and specific areas to improve, not just a vague score. Sessions are tracked over time so a user can see real improvement, whether their pacing is steadying or their filler words are dropping, session over session.
How we built it
On the backend, we used Supabase to handle our database and user accounts, paired with Google Cloud for authentication, so users can sign up and log in securely and pick up their session history wherever they left off. For the core feature, a recorded answer gets processed so the platform can understand the actual words being said, then that transcript is run through Grok, which evaluates tone, pacing, and filler word frequency and turns that into a personalized feedback report a student can actually act on.
Challenges we ran into
Two things gave us the most trouble. The first was getting the AI to reliably understand what was actually being said and then figuring out what that should mean as feedback, turning a raw transcript into specific, useful advice instead of a generic score took a lot of iteration. The second was building out login and sign up, wiring together Google Cloud and Supabase for authentication took longer than we expected, since we had to make sure user sessions and account data stayed in sync and secure across both services.
What we learned
Building the sign in flow taught us a lot about connecting different services together. Getting Google Cloud and Supabase to work with each other for authentication meant learning how account data and sessions needed to pass between the two, and that ended up teaching us more about how real world authentication systems fit together than either of us expected going in.
What's next
Our next priorities are video and eye contact analysis to give feedback on body language alongside voice, an expanded question library covering technical, case style, and leadership interviews, personalized coaching that spots patterns across a student's sessions over time, and mobile access so practice can happen anywhere, not just at a desk.
Built With
- ai
- canva
- claude
- gpt
- supabase
