A graph of what I've spent my time on.

Like others, I came across Andrej Karpathy's gist on building a personal wiki with an LLM. The idea was simple. Treat your own history, notes, transcripts, documents as the corpus, and let a model help you navigate it the way you'd navigate Wikipedia. A personal context graph, built from the stuff only you have.

I became obsessed with this idea and intrigued based on mixed results from using BigQuery as a vector table and RAG approach. I already had the corpus. Every client meeting I've taken over the last three years is recorded and transcribed through Tactiq. That's over two thousand meetings of real client work: data platform migrations, team hires, model evaluations, stakeholder debates, decisions made and unmade. It's the best proprietary data I have for writing about this work honestly, because it's the work itself.

Each meeting became a markdown page with a short summary and a set of wikilinks to the themes it touched. Topics like data governance, identity resolution, stakeholder alignment, AI agent design, and team building emerged as recurring nodes. Over time the links formed a graph that reflects what I've spent my time on.

Below is a representation of the graph I came up with. Further down the page, I've written about how I use it to find what to write about, and the pipeline I've built around it.