The log

NOTES ON
BUILDING AI

Essays on building AI for real businesses and government — architecture, local LLMs, knowledge, prediction, and meaning before automation. No hype, only systems that work.

Machine Learning Prediction Models for Real, Messy Business Data Why most prediction models fail on actual Greek business data and how to build ones that work with incomplete spreadsheets and manual entries. July 5, 2026 Read → AI Software Architecture What It Really Feels Like to Work as an AI Software Architect What the job of an AI Software Architect actually looks like — messy files, real workflows, responsibility, and knowing when to say no. July 2, 2026 Read → AI Knowledge · OKF OKF: Simple Structure for AI Knowledge The problem in AI projects is often not the model but messy knowledge. OKF (Open Knowledge Format) is a simple way to structure it. June 29, 2026 Read → Local LLM · Sovereignty They Turned Off a Frontier AI Model Overnight. Still Think the Cloud Is "Yours"? A frontier model was switched off overnight. The case for local LLMs: own your floor, keep the cloud as your ceiling. June 27, 2026 Read → AI Architecture AI Software Architecture: The Part People Don't See AI Software Architecture is not the impressive demo. It is understanding the real problem, the workflow, the risk — then the technology. June 25, 2026 Read → AI Prediction AI Prediction Models: The Art of Seeing Patterns Before They Happen Comparing forecasting models for sales data — Seasonal Naive, Croston, SBA, LightGBM Tweedie. Each has its own personality. June 25, 2026 Read → AI in Government Meaning Before Automation: AI in Government and Business What early-stage AI in government and business really needs: understanding information before automating it. June 9, 2026 Read →