I used gpt-4o-mini for image alt text generation and gpt-4.1 for other suggestions. I built by adding features one by one with agent and assistant, spent quite a lot of time on tweaking the analysis suggestions.
Seems there is a bug with the page limits, I didnt test that so much.
Pretty much refactored the whole analyzer pipeline and added features. You can now chat with AI about the suggestions and retrieve page content for context etc. I think its pretty solid now.