Mind Mapper

Announcing Mind Mapper: From Markdown to Visual Maps

Transform your markdown documents into interactive mind maps. Mind Mapper converts headings and subheadings into visual diagrams with rich tooltips for additional context. Export your results as .svg, .png or .html for easy sharing and embedding.

After getting the mind map, you can drag the nodes of the mind map to new positions.

Try it Here

Spike Solution - Tree Formats

Originally I conceptualised this as taking bulleted lists to mind maps. Later I realised that the actual value of the mind map was in the notes attached to each node. I needed to implement that part. So in turn, instead of thinking in terms of bulleted lists, I should think in terms of markdown docs with headings and subheadings.

But whilst I was still thinking in terms of bulleted lists, I was seeing a need for conversions of different ways of representing a tree with text. I needed a converter that will take a tree, as text, and convert it to other tree formats. I asked Claude to support ASCII art, bulleted lists and lists of paths.

The first attempt at this was a wash, with Claude only parsing the first line - and failing to correct the fault when asked to. So I prompted again, and as sometimes happens with a fresh start got a decent spike solution.

The auto-detection of format failed for the tree of exercises, as it turned out Claude was treating anything wait a '-' in it as indicating a bulleted list. Push-up had that pattern! So I rearranged the tests, and tweaked that part. Could be lots better at detecting, but this is a spike solution. It's good enough.

The code for generating the ASCII art is actually a bit dodgy. Give it a more deeply nested tree than its test examples and it will start duplicating nodes. It works for the test cases.

Data

Simple prompts gave me more extensive trees than these examples for topics like exercise, meditation, molecules in biology. It became clear it is very easy to generate topic trees. Tip: After getting the topics, run an AI fact-check on them in another session. This may catch some errors if the fact checking engine can do web searches. I decided that for now I'd actually keep the trees small - punting the problem of dealing with mind maps too large to fit in the area to later. So I prompted Claude to write a paragraph on each of the topics in the original two trees.

I liked what Claude made. Gemini's fact check gave clarifications about sencha tea harvest vs steeping, that oxidising tea was deliberate, not accidental - even though it makes a less good story. Gemini noted that lower body muscle mass is nearly 50%, not roughly 60%, and that bicycle drivetrain is efficient but the overall conversion of chemical energy to movement isn't really. Few corrections, and nothing seriously off.

I got Gemini to propose new wording, rather than just point to the errors for me to fix. It is quite clear that this kind of curation of nuggets of information can scale easily. It would be easy to prompt to get interesting 'cards' on many topics. LLMs are good at these essentially factual sound-bites. It also points to the possibility of dynamically extending the tree. With an LLM on the other end, you could drill down into a topic, and go from a tea topic to a new tree on trade routes or aromatic plants.

Chronology