When I was making Quizzine, I faced a big issue. I had the NPTEL quiz data (The question statements, the options and the correct option index) as PDFs, but I struggled to arrange them as JSON so my app would be able to parse and display it in app. I used quite a number of tools and it was quite painstaking to do this.
Ultimately I ended up using a weird combination of AI and Find and Replace to get the job done, but I wish it was much more streamlined than this. Multimodal AIs have the full ability to understand PDFs. They just lack the ability to properly read/write/edit a high context file like a JSON data store. The JSON file storing the quiz data for 2 NPTEL courses was 2111 lines in my case! At one point, Cline gave up and garbled up all of my data no matter what I did. That really sucked.
There needs to be a more efficient way to do this than to ask Cline. I envision a smarter AI tool which doesn’t pass in the entire file and hope for the best. I want this AI to use features like find and replace to be able to find the fields that require the edits and make them. I want this AI to not edit the fields its not supposed to, which means just not giving those fields as context to the AI in the first place.
Seems pretty easy to build, and will make a very wanted launch right after I FOSS Quizzine.