Slide Doctor

Inspiration

The AI tool that helps you when it's 8 a'clock in the morning and you have to show a PPT slide-deck at 9 is missing.

What it does

Slide Doctor will make Powerpoint a breeze. Now everyone can have hi-quality, professional looking PPT slides. It aims to be the ultimate Multi-modal A.I. tool, powered by Pixtral 12B for fixing PPT Presentations.

How we built it

• use open source libs to convert said slide(s) to PDF(s) then from pdf to PNG(s)
• use the a set of A.I. agents, including Pixtral 12B LLM and analyze the image(s)

Challenges we ran into

Problems with the net stack when using the Mistral API

Accomplishments that we're proud of

We were able to read a slide deck both with AI computer vision and trough reading the PPTX file itself using open-source tools

What we learned

How to use Mistral and how to use different API tools.

What's next for Slide Doctor

Ideally, the tool should:

• markdown to slide // create exec summary?
  • compile all titles and slide images
  • https://github.com/MartinPacker/md2pptx
  • align with LLM to be the same style (unzip pptx and show the slide xml vs another slide xml and tell llm to be consistent)
• fix pptx
  • do it on a copy, is it valid? if it is, replace the original
• front end for showing errors - gradio
  • show image of the page and list issues
  • tick the once to solve
  • one page per view
• audio dictation: where the user will activate his/her microphone and dictate what the PPT slide deck should contain then a fleet of agents should generate the slides ( images with Stable Diffusion, text, charts, etc. )

Built With

• mistral
• pip
• python
• uv