I have been using NotebookLM to speed up my reading lately. Summaries, highlights, curated clips.
It is faster, sure. But something was off.
The slow part where you sit with an idea and work out what to do with it, was handled by AI.
That’s thinking.
And thinking is where the actual learning happens. Real world application forces you back into that seat. It is uncomfortable, but it is also where things start to stick.
I built Munger Vault to study Charlie Munger’s mental models. Here’s what I did:
Create a strict boundary. I uploaded Munger’s speeches, essays, interviews, and videos into a NotebookLM notebook. The AI stays inside that source material and does not hallucinate.
Build a simple front door. I used Streamlit to create a minimalist dashboard with a single text box. I type a real-world dilemma, and it returns Munger’s diagnostic framing.
Connect the plumbing. I use Google Antigravity to prepare a short Python script. The script acts as the bridge. It takes my scenario, passes it to NotebookLM, and returns clean output.
Capture the output. I save insights and my own notes into Obsidian. The conversation becomes a permanent intellectual asset.
I am not just reading about Munger anymore. I am running problems past him. I have since run the same workflow for the Bible, Buddhism, and Stoicism.
I continue thinking and reviewing my notes. How should I start using them more and collecting less. Something to watch.
Just another coffee thought from Gav.
