AI Boosters: Don’t worry, AI will create LOTS of jobs!

The jobs:

Since that wasn’t an option here, I just specified the output format directly in the prompt, and it turned out easier to validate the output manually in Obsidian based on the filename

the script source code: github.com/dmitriiweb/r...

At first, I planned to use gemma3 for image recognition, but it doesn’t support tools, so I had to wrestle with the prompt—and even then, some manual cleanup is required. If the model supports tools, you can define a pydantic.BaseModel in pydantic-ai with the required arguments and types (!) ->

So I wrote a small script that converts all PDFs into images, then passes each image through a multimodal model, and saves the result as Markdown with proper formatting ->