AI Voice Interview Coach

A voice-based AI interview coach that conducts realistic mock interviews using your resume, gives real-time feedback, and helps you practice for the real thing. Talk to it like a real interviewer — it listens, responds, and coaches you on the spot.

How It Works

• Pick an interviewer persona (recruiter, technical, hiring manager) and optionally a target position
• The AI reads your resume and conducts a live voice interview over WebSocket
• Speech-to-text (Deepgram Nova-3) transcribes you in real time
• Claude generates interviewer responses, streamed as text and spoken back via TTS (Deepgram Aura-2)
• Session transcripts and coaching notes are saved as markdown files

Data Setup

Everything runs off markdown files in data/. Set this up first — without a resume, there's nothing to interview you on.

data/
  candidates/{your-name}/
    resume.md              ← your resume (required)
  interviewers/
    recruiter.md           ← included: warm, behavioral focus
    technical.md           ← included: system design, engineering depth
    manager.md             ← included: leadership, team fit
  positions/
    your-target-role.md    ← job description (optional, tailors questions)

To get started: create a folder under data/candidates/ with your name and add a resume.md with your experience, skills, and background. Plain markdown, no special format required.

Interviewer personas are included. Positions are optional — add a markdown file with the job description and requirements to get role-specific questions.

Prerequisites

• Bun (v1.0+)
• A Deepgram API key with Member role (not just API-only — Member role is required for streaming STT)
  • New accounts get $200 in free credits at deepgram.com
• An Anthropic API key — or a Claude Pro/Max subscription (use your auth token from claude.ai)

Setup

git clone <repo-url>
cd voice-interview

# Install dependencies
cd backend && bun install && cd ..
cd frontend && bun install && cd ..

# Configure environment
cp backend/.env.example backend/.env
# Edit backend/.env with your API keys:
#   DEEPGRAM_API_KEY=your-key-here
#   ANTHROPIC_AUTH_TOKEN=your-key-here

Running

# Terminal 1 — backend (port 8081)
cd backend && bun run dev

# Terminal 2 — frontend (port 5173)
cd frontend && bun run dev

Open https://localhost:5173 in your browser. Grant microphone access, pick your interviewer and position, and hit Connect.

Stack

Layer	Tech
Runtime	Bun + TypeScript
STT	Deepgram Nova-3 (WebSocket streaming)
LLM	Claude via Anthropic SDK
TTS	Deepgram Aura-2
Frontend	Vite + React + Tailwind
Storage	Markdown files on disk

License

MIT