Vocalize — Interview Companion

An entirely local, voice-first AI mock interviewer to practice technical communication

Quick Setup (Windows PowerShell)

Create and activate a virtual environment, then install dependencies:

python -m venv .venv
.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Create a .env in the project root for any environment variables you need:

# .env (example)
# Add any keys your app or Modal needs, e.g. API keys
# EXAMPLE_KEY=your_value_here

Run (local development)

Start the Gradio app:

python app.py

Gradio will print a local URL (and a public share URL if enabled) where you can open the UI.

Notes:

• app.py expects a running Modal backend referenced via the modal client (the code uses Cls.from_name to obtain the remote class). For pure local development you can:
  • Mock modal_model with a lightweight local implementation that exposes transcribe.remote() and generate.remote() methods, or
  • Replace those calls temporarily with local stubs to test the UI and audio handling.

Modal backend

• modal_backend.py defines a Modal App and the CodingCompanion class which loads a language model and tiny Whisper for transcription. Deploying this requires Modal credentials and following Modal's deployment docs.
• If you plan to deploy to Modal, ensure you have Modal configured locally and the GPU/image requirements met.

Audio handling

• The Gradio Audio component uses type="numpy" and soundfile for conversions. Ensure soundfile is installed and functional on your platform.
• Audio transcription flow:
  • Gradio provides a (sample_rate, numpy_array) tuple.
  • app.py writes the numpy array to an in-memory WAV via soundfile and sends bytes to the backend for transcription.

Testing

Run tests (if you use pytest):

python -m pytest -q