EV Virtual Assistant

An initial Python project scaffold for building a virtual assistant named EV with webcam, microphone, and fine-tuning support.

What this scaffold includes

• main.py: basic webcam capture, microphone recording, wake-word detection placeholder, and model query flow
• fine_tune.py: dataset preparation and example fine-tuning submission function
• dataset_utils.py: helper for building a JSONL training dataset
• requirements.txt: Python dependencies for audio/video capture and OpenAI interaction

Setup

1. Create a virtual environment:

python -m venv .venv
source .venv/bin/activate

2. Install dependencies:

pip install -r requirements.txt

3. Create a .env file with your API key:

OPENAI_API_KEY=your_api_key_here

Running the assistant

python main.py

This will open the webcam and listen for a wake word like ev in the recognized speech.

Fine-tuning workflow

1. Add training examples in dataset_utils.py or your own dataset file.
2. Run python fine_tune.py to export data and submit it.
3. Update main.py to use your fine-tuned model once ready.

Notes

• This scaffold provides placeholders for voice recognition and speaker recognition.
• To recognize you by voice, you will need a speaker identification model or a voice print dataset.
• For reliable wake-word detection, use a dedicated wake-word engine or training dataset rather than simple text matching.