Live
Voice-to-Graph
Send it a voice memo from your phone — an AI agent splits your spoken ideas apart and links them into a topic graph. All your ideas in, book material out.
I think out loud. Voice-to-Graph turns that into an organized, cross-linked knowledge graph — automatically.
The pipeline: I record speech — a thought on a walk, a full hour-long lecture, an argument with myself. It’s transcribed, then an AI curation service reads the transcript, splits it into individual ideas, decides where each one belongs in my Obsidian vault, places it, and cross-links it to related notes it already knows about. Many ideas, recorded over months, weave themselves into one connected map of thinking.
Under the hood it’s a self-hosted stack I built end-to-end: Obsidian running in the browser, multi-device sync via CouchDB, and a FastAPI ingest service that calls the Claude API to do the placement and linking.
The part I find most satisfying is the five integration surfaces — the same brain is reachable from the browser, from the sync client, from Claude on desktop and phone through an OAuth MCP server I wrote, from ChatGPT through GPT Actions, and from iOS Shortcuts. Build the app once, then wire it into every AI assistant.
For content creation it’s become a quiet superpower: every idea I’ve ever recorded is already filed, linked, and searchable.
Want someone who builds like this?
I'm open to opportunities. Tell me what you're building — I answer every email.