Multimodal Structured Extraction with Gemini 2.0 Flash and Google GenAI Python SDK 1.0

In this tutorial we extract structured data from a variety of sources, including an investment newsletter PDF, a podcast, and a YouTube video.

Setup

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Create a .env file in the root directory with your Gemini API key

GOOGLE_API_KEY=your_google_api_key

Handy Commands

Download YouTube video

yt-dlp https://www.youtube.com/youryoutubeurl

Download YouTube audio

yt-dlp -x --audio-format mp3 https://www.youtube.com/youryoutubeurl

Split audio

ffmpeg -i input.mp3 -f segment -segment_time 10 -c copy output_%03d.mp3

or use the shell script split_audio.sh

./split_audio.sh input.mp3

Extract audio from video

ffmpeg -i input.mp4 -q:a 0 -map a output.mp3