AIMM Hub

Mastermind recap

Cognitive Twins and Local-First Knowledge

· AIMM · 90 min

Facilitators: Lou D'Alo

AI Mastermind | Knowledge Entrepreneurs Edition

“The goal isn’t a chatbot that sounds like you. The goal is a system that decides like you. That’s the difference between building a golf cart and building a Ferrari.” — Lou

This Week in 30 Seconds

  • Anthropic’s Compute Crunch — Why Opus 4.7 feels worse, the hidden 50% tokenizer tax, and Lou’s case for staying the course instead of platform-hopping.
  • The Local-First Stack — Why your knowledge should live on your disk, not in a platform — and how CLAUDE.md + Obsidian + Claude Code form the right foundation.
  • The Cognitive Twin — Lou’s DSPy-powered system that mines your daily AI conversations to build a model that replicates your decision-making, not just your writing voice.
  • Code vs. Inference — The underestimated efficiency lever: Scott ran 90 million records down to 128K with Python + Gemini; Lou’s Gears build went from 6 hours to 15 minutes with the same principle.
  • GEARS Wins — Don’s site landed a discovery call from Ghana the day after launch, and multiple members are in the final stretch of integration.

The Cognitive Twin: Teaching AI to Decide Like You

There’s a difference between an AI that sounds like you and an AI that decides like you. Lou is building the second one.

The centerpiece of this week’s session was a 6-minute NotebookLM-narrated video summarizing what Lou has been quietly developing: a cognitive twin architecture built on DSPy, a Stanford framework originally designed for programmatic AI optimization — and repurposed here for something considerably more personal.

The insight that unlocked the project: every AI interaction is secretly generating high-quality training data. Every time you correct Claude’s framing, override a suggestion, redirect a conclusion — that’s a decision instance. A before-and-after. A data point that captures not what you said, but how you think. The cognitive mirror skill harvests this in two modes: mine mode scans conversations for the principles behind your decisions (the structural how), and harvest mode captures specific correction moments as clean before/after training pairs. Both run on automated daily and weekly schedules. You just do your normal work.

Once you accumulate roughly 40-50 decision instances, DSPy can work. Instead of you writing a prompt to produce an output, DSPy analyzes your golden examples and generates the prompt that has the highest statistical probability of reproducing your caliber of result. It runs iterations. It runs tournaments. “Instead of you trying to write the perfect prompt, the optimizer goes and tests thousands of different instructions against your data to find the one that works best. It does the hard work for you.”

The junior associate analogy Lou returned to: “You don’t hand them a 500-page rulebook. You give them a stack of 50 of your completed projects and say, go study these until you can tell me exactly what I do on the 51st. That is exactly what DSPy is doing.”

Next session: Thursday, April 30, 2026