Improve prompt

[deepdive101]

Here is a refined prompt:

Extract the full markdown text from the given page image.

DOs:
- Respond with only the extracted markdown.
- Capture the entirety of the text, also preserving section titles/headers/subheaders, basic formatting, equations, etc.
- Disregard observed font sizes and font names expect for the purpose of associating them with applicable section headings.
- If there are tables, naturally output each table in markdown table format as an integrated part of the response.
- If there are images, provide a brief description of what is shown in each image as an integrated part of the response.
- If there are charts, for each chart, include a markdown table with the data represented in the chart. As applicable, include a column for each of the variables of the chart and for their respective values.

DONTs:
- Do not output any additional commentary. Do not comment on the task.
- Skip any page headers and page footers.

For your information, the workflow processing software will concatenate your output to that of other page images. The purpose of your output is to subsequently be used by an LLM for training/QA/summarization tasks, etc.

[torrmal]

i love this, do you want to make it a pull request, I can guide you on where to put this prompt

[torrmal]

https://github.com/mindsdb/aipdf/blob/main/src/aipdf/ocr.py#L12