I inherited a really crappy comparison table which has a lot of really excellent data in it. This thing is about 3K of HTML with all the cells presented as divs, which is probably why GPT-5 and Gemini Flash can’t make sense out of it.
So…….would a vision model possibly fix this? I think there are some vision models in the current selection but vision is being blocked, maybe?
Anyway, I am planning to do an AI recognition of the image of the table, and feed that into the KB.
Is there any simpler/better solution?
1 Like
You can try uploading an image of the table directly, and the chosen AI model will do its best to extract and interpret the data.
If possible, it may also help to reformat your data into a more structured or machine-readable layout (for example, converting nested divs into a standard HTML table or exporting it as a CSV). That gives the AI a clearer understanding of rows, columns, and relationships between cells.
Additionally, consider incorporating detailed instructions in your prompts that clarify the table’s structure, formatting, and the interpretation of its cells. Supplying this contextual information can enhance the model’s ability to process and understand the data accurately.