Ask HN: As an eng leader, how useful is AI-generated internal documentation?
Hi HN!
This past weekend, I built a tool called GitSummarize, it's an open-source ai-powered documentation tool.
Given any public GitHub repository it generates high-quality documentation from the codebase using Gemini 2.5 Pro. Specifically, it tries to extract the tribal knowledge, business logic/rules embedded within codebases.
Pro Tip: You can replace "hub" with "summarize" in any repository URL to access its documentation.
I created this because there was a steep learning curve to understanding some of the open-source codebases out there while getting started with contributing. Having detailed documentation would have been super helpful in those situations.
This project was heavily inspired by GitIngest and GitDiagram so make sure to check those out as well!
Give it a try and I would appreciate any feedback! Would also love to know how useful this would be to developers or engineering leaders for their teams?
Nice work, but... people do not want documentation, they want to ask someone that knows both the software they do not know and they want someone that knows what they are doing already, so the developer gets simply told what to do without having to think or do research on their behalf. Yes, this is lazy ass, this is dangerous, but it is how people behave.
I'd think something as dumb (for the user, but complex to make) as a GraphRAG system that already has a developer's code repo and design docs in the system, and is then able to be given "candidate libraries" which the GraphRAG+LLM reasoning agent then evaluates the candidate against the active project for compatibility. That would be one decent application, but the better one is once a "candidate library" is accepted, that GraphRAG+LLM agent can just-in-time explain to the developer how a library works and should integrate with their active code base than any documentation.
This is super amazing. I also go the "Failed to ..." etc error messages initially ( waited about 3-4 mins that time). Just checked once more after about 1 hour and it did generate the whole documentation.
Fairly readable and doesn't seem wrong ( havent gone in depth yet)
https://gitsummarize.com/daptin/daptin
Tried to look at one of my open source repos and kept getting the error “Failed to process codebase. Please try again later or try adding your own Gemini API key.”
Edit: fwiw I am an “engineering leader,” and I do think it would be nice to have a model that could answer questions about our codebase at work, but we’d need some guarantees that the data would not be used for training, stored inappropriately, etc.
Where is the link, I am coming to use cause want to learn from a project