Hacker News new | ask | show | jobs
by vunderba 672 days ago
You can actually do this now - I've taken a great deal of my hand written workflow diagrams, snapped a picture of them, and asked chatGPT to convert them to mermaid UML.

It was one of the more impressive demonstrations of multimodal AI.

You can also import mermaid into Google's Draw.io app if you want to reflow or change the theme.

3 comments

I did something really similar by converting handwritten math to LaTeX.

This use case (and yours as well) can be really useful and works well just by using ChatGPT. In many cases people are just building wrappers of gpt-4o, but I can see some room for improvement by building on top of OpenAI’s APIs.

Even better - Excalidraw supports importing mermaid - and it has a built in AI tool where you can bring your own LLM via an API to generate diagram without even leaving the app.
an FYI-You-should-too -- I am a huge fan of telling claude to give me mermaids and swim diagrams of most of the logic and code I am iterating with.

I like to do the following:

"Review this code and give me a readme, include mermaid, swim diagrams to draw the logic."

Claude is great because it renders the mermaid in the artifact (most times).

then I can copy the mermaid into a mermaidlive (I still havent found the best mermaid site because some of them choke on the mmd output of claude and I dont know how to fix it sometimes...

for example - here is a mermaid where I was looking into how the universities, professors and families of SBF, Holmes and Ellison were tied:

https://i.imgur.com/njFKvmJ.png

Here is a mermaid of who sits on the agra committees, who funds their campaigns, and who did they make investments into wrt to the sugar lobby:

https://i.imgur.com/FGGcvxL.png

Having chatbots do mermaids is fn fantastic