Personal project · AI · Knowledge management
Second Brain
My knowledge base: a Markdown wiki maintained by AI from my documents, with a local app to explore it as a graph.
Overview
It brings all my projects' information together in one place. I add the sources—documents, spreadsheets, notes—and Claude Code processes them: it summarises each one, creates pages for people, companies and topics, links them together and flags contradictions and missing data. A local app lets me explore it with search, backlinks and a connection graph.
The challenge
With several projects at once, information was scattered across folders, chats and emails, and every question to an AI started from scratch.
What I did
- Defined the system's rules: sources are never modified, one page per idea, every fact traced to its source and flags for contradictions, pending items and inferences.
- Designed the workflows: ingesting a source, querying and health-checking the wiki, with one commit per operation.
- Built the local app: a Node.js server with Express, an interactive graph, backlinks and automatic reload whenever the wiki changes.
- Separated what can be shared from what is confidential, with rules that keep sensitive data out of the wiki.
How it works
Untouched sources
Originals are never edited: the wiki summarises them and cites the source of every fact.
Links and graph
Every page links to the others, and the graph colours nodes by project and enlarges the most connected notes.
Quality flags
Contradictions, missing data and inferences are flagged so a document's content is never confused with what the AI infers.
Live reload
The app refreshes itself while the AI edits the wiki.
Stack and tools
- Claude Code
- Markdown + YAML
- Node.js
- Express
- vis-network
- WebSockets
- chokidar
- marked
- gray-matter
- Git · GitHub
Skills
- Information system design
- AI agent workflows
- Documentation
- Data governance
What I learned
- Clear rules for the AI—what it may touch, how to cite and when to ask—are worth more than any one-off prompt.