https://christopherclaudecook.com/insights Writing on production AI, AI leadership, and secure language-model deployment. en-us https://christopherclaudecook.com/insights/inside-cognicore-memory-trust-distillation https://christopherclaudecook.com/insights/inside-cognicore-memory-trust-distillation Thu, 08 Oct 2026 12:00:00 GMT The mechanics under a local AI companion: how three incompatible signals become one memory ranking, how a 27-billion-parameter model teaches a 0.6-billion-parameter one each night, and the skill score that decides how much the small model is allowed to steer. https://christopherclaudecook.com/insights/finding-people-with-radios-and-radar https://christopherclaudecook.com/insights/finding-people-with-radios-and-radar Mon, 05 Oct 2026 12:00:00 GMT Two home projects that tried to locate people from radio signals: router signal strength first, then a mesh of ESP32 boards using radio tomography, then a small mmWave radar in each room. How each sensor works, what the simulations say, and what none of them can do. https://christopherclaudecook.com/insights/skyready-107-part-107-study-app https://christopherclaudecook.com/insights/skyready-107-part-107-study-app Sat, 03 Oct 2026 12:00:00 GMT An offline study app for the FAA Part 107 Remote Pilot knowledge test, with a cited source on every answer and a mock exam that mirrors the real one. Plus an interactive look at load factor, the piece of flight physics the exam cares most about. https://christopherclaudecook.com/insights/local-ai-with-its-own-memory https://christopherclaudecook.com/insights/local-ai-with-its-own-memory Thu, 01 Oct 2026 12:00:00 GMT CogniCore runs on one 24 GB graphics card, keeps a long-term memory, and retrains a small model every night. The rule that decides when those predictions are trusted, and the voice pipeline that lets you interrupt it mid-sentence. https://christopherclaudecook.com/insights/ethos-knowledge-base-mcp https://christopherclaudecook.com/insights/ethos-knowledge-base-mcp Thu, 27 Aug 2026 12:00:00 GMT A language model starts every session with an empty context window, and vendor memory features solve that inside one product while fragmenting it across the rest. How a self-hosted vault attached to both ChatGPT and Claude handles retrieval, artifacts, and a semantic memory the user can watch grow. https://christopherclaudecook.com/insights/scaling-ai-team-1-to-10 https://christopherclaudecook.com/insights/scaling-ai-team-1-to-10 Sun, 23 Aug 2026 12:00:00 GMT I founded ORBIS's AI department as its only engineer. Five years later it was a core team of ten inside programs of 30+ people spanning company departments, contract experts, the U.S. Navy, shipyards, a university, and industry partners. The hurdles, the results, and the servant-leadership model that held it together. https://christopherclaudecook.com/insights/lidar-driver-from-the-serial-protocol-up https://christopherclaudecook.com/insights/lidar-driver-from-the-serial-protocol-up Mon, 10 Aug 2026 12:00:00 GMT A pure Node.js driver for the RPLIDAR A1M8 with no vendor SDK: decoding standard, express, and delta-compressed boost scans, testing against a byte-level mock device, and a simple scan-matching mapper. Includes an interactive version of the mock room. https://christopherclaudecook.com/insights/production-ml-lessons-from-defense https://christopherclaudecook.com/insights/production-ml-lessons-from-defense Sat, 01 Aug 2026 12:00:00 GMT Most ML projects fail in the distance between a validated notebook and a system running inside someone else's workflow. What years of building AI for Department of Defense inspection pipelines taught me about that distance. https://christopherclaudecook.com/insights/agent-towns-workhorse-and-busyworld https://christopherclaudecook.com/insights/agent-towns-workhorse-and-busyworld Thu, 18 Jun 2026 12:00:00 GMT Early design notes from an experiment still in progress: several AI agents in one shared simulated world, connected like multiplayer clients, with the world server holding the only copy of positions and time. https://christopherclaudecook.com/insights/local-llms-secure-environments https://christopherclaudecook.com/insights/local-llms-secure-environments Wed, 20 May 2026 12:00:00 GMT For defense, healthcare, legal, and finance workloads, sending data to a commercial API is often prohibited outright. What actually works when the models have to run inside the boundary.