Blog
Notes on software control, vendor risk, and prototype readiness
Practical writing on software projects, vendor evaluation, maintainability, workflow design, AI-built prototypes, and realistic automation.
Topics
Browse by topic
How I structure AI-assisted development: slides from MAINDSET Žilina
AI coding becomes useful when the work around it is designed just as carefully. These are the slides and resources from my MAINDSET talk in Žilina.
I wanted to understand a perceptron, so I built one
I wanted to understand what a perceptron actually learns, so I used AI to build one I could train, break and inspect. The demo became a much better lesson than another explanation.
I turned podcast transcripts and papers into an Obsidian second brain
I used Karpathy's llm-wiki pattern, Obsidian and GPT-5.6 Sol High to turn podcast transcripts and research papers into a maintained knowledge base, then used it to produce evidence-aware sleep and nutrition guides.
Stop Outsourcing AI Thinking
Most SMEs start AI adoption by shopping for tools or automation. A better start is to help the people who already understand the business use AI well, then automate only what proves stable.
The First Coding Benchmark That Feels Like My Actual AI Work
DeepSWE separates coding agents in a way older benchmarks often did not. That matches what I have seen using Codex and Claude side by side: the best tool is the one that survives real work, not the one with the loudest hype cycle.
Need someone technical on your side before software gets expensive?
Bring me the proposal, project, or prototype. I'll tell you what looks reasonable, what is risky, and what to decide next.
I reply within 1 business day.