I built a self-maintaining knowledge base for work. Meeting notes, brain dumps, initiatives, and issues compile automatically. Notes link to people, partners, employees, vendors, and clients. Notes also link to initiatives and issues. A dashboard shows emergent patterns and problems across all of it.
My job is to feed it, make the occasional correction, and ask it questions.
Every query gets added back as a note. So if I ask about something today and take action on it next week, the system connects those dots. It knows what I asked. It knows what I did.
I started with Karpathy’s LLM-wiki tweet. Then I had Claude read nvk’s version and pull in what fit. The schema that came out of that was solid, but it was built for research. I needed something different.
I told Claude my use case was professional. It assessed the schema and changed the orientation. I needed to track decisions, issues, and initiatives. I needed Claude to remember my rationale, flag emerging problems, and track progress. So I asked it to fix the schema.
That was a week ago. The schema will probably change again. Claude will do most of that work too.
I just keep feeding it.