I like the idea and know a lot of people who would profit from something like that, however for me sadly it only works for really basic charts. And scaling is not consitent. E.g the scale in the pic goes from 500 to 530 and the ai scales it from 0 to 600 where you cant see anything anymore. The good thing is AIs are getting better and better any may tomorrow the problem is solved.
+ Do you hand the pic right to Gemini Flash or do you have middle steps to extract the scales and so on?
Thanks a lot for your reply and constructive Feedback, we really appreciate it! Yeah, right now it has a hard time with complex charts and works better with simpler versions. There is a lot of improvements that could be done here. For now we wanted to see how useful the community finds it and what the most common use cases are before making further improvements. ChartAI was built in just a week and is very much a MVP to further improve. Regarding your question: Yes, we directly hand it to Gemini Flash, however, we have a multi step prompting logic in place. Basically, we first ask the model to extract the data in a certain JSON format that worked best during testing. Then the model is asked to convert this JSON to chartJS code, where we also provide the model with code context. Then some code cleanup is done outside the mode in JS. If you have any further thoughts or questions, please let us know
+ Do you hand the pic right to Gemini Flash or do you have middle steps to extract the scales and so on?