VoxInterview – A Voice-First AI Interview Coach

VoxInterview Thumbnail

Inspiration

Interviews are stressful — not because people lack skill, but because they struggle to communicate that skill clearly under pressure. I wanted to build something that feels less like a test and more like a coach. VoxInterview was inspired by the idea that practice should feel natural, conversational, and adaptive — closer to talking with a real interviewer than filling out another form or reading canned questions.

As someone who’s spent a lot of time preparing for technical interviews, I wanted a tool that could simulate real interview pressure, give meaningful feedback, and help users improve how they communicate their thinking, not just what they know.

What it does

VoxInterview is a voice-first AI interview coach that simulates realistic technical interview sessions.

Users:

Choose a role (Frontend, Backend, or Security)
Hear spoken interview questions using text-to-speech
Answer verbally using their microphone
Receive AI-powered feedback analyzing:
- Clarity
- Confidence
- Relevance
- STAR-method structure (Situation, Task, Action, Result)
Get a confidence score and actionable improvement tips
Progress through multiple interview questions in a session

The experience feels like a real mock interview — not a form, not a chatbot, but a conversation.

How we built it

VoxInterview is built as a full-stack web application:

Frontend

React for the UI
Audio recording via the Web Audio API
Real-time playback of AI-generated voice questions
Visual confidence scoring and structured feedback display

Backend

Node.js + Express API
ElevenLabs for speech-to-text and text-to-speech
Google Gemini for intelligent response evaluation
Custom STAR scoring system to assess interview structure
Circuit breaker logic to gracefully handle AI outages

Architecture Highlights

Role-based interview logic (frontend, backend, security)
Modular scoring and feedback engine
Fallback systems when AI services are unavailable
Deployed as a standalone API that the frontend consumes

Challenges we ran into

Balancing realism with reliability — AI responses can be unpredictable, so we built fallback systems to ensure the app never “breaks” mid-interview.
Audio handling — managing recording, transcription, and playback cleanly across browsers required careful timing and error handling.
Meaningful feedback — translating raw AI output into structured, useful coaching took several iterations.
Deployment complexity — coordinating frontend hosting, backend APIs, environment variables, and file handling on a shared server took careful planning.

Accomplishments that we're proud of

Built a fully voice-driven interview experience end-to-end
Implemented a STAR-method scoring system automatically from spoken responses
Designed a system that gracefully degrades when AI services fail
Created a product that feels genuinely helpful, not gimmicky
Deployed and running live on a custom domain