The World Model Equation Is a WHERE Clause
The first code post in this series opened with a jab at explanations that lead with “a conditional probability, a subscript, a latent variable, and half …
Ai
A World Model in Forty Lines of TypeScript
Every explanation of world models I’ve read starts the same way. A conditional probability, a subscript, a latent variable, and half the audience gone by …
Your Time Series Is Already Training Data
At Unimo we spend our days watching telemetry from energy hardware. Temperatures, states of charge, power flows, a reading every few seconds from devices …
Compute Isn't the Constraint
After When the API Gets Switched Off went up, Bram De Buyser, an old friend, wrote back and put his finger on a line of thought I’d neglected. The …
When the API Gets Switched Off
On 13 June, Anthropic disabled Claude Fable 5 and Mythos 5 for every customer outside the United States. Not deprecated, not rate-limited, but disabled. US …
Streaming AI Responses from Server to Screen
I wrote previously about a partial JSON parser for extracting text from incomplete LLM responses as they stream in. That post covered one specific piece: …
What It Actually Takes to Fine-Tune an LLM in Europe
This series has so far dealt with deployment. Where to host models, how to wrap them in compliance logic, which providers can serve inference, and how to verify …
Securing the Open Source AI Supply Chain
This is the fourth post in a series on European AI sovereignty. I have covered why jurisdiction matters, how to handle UK-EU regulatory divergence in code, and …
A Practical Guide to the European Sovereign Inference Stack
I wrote previously about why jurisdiction matters for European AI platforms and then about how to handle the UK-EU compliance split in code. Those posts covered …
Architecting for Two Europes with Compliance-as-Code
I wrote recently about building AI platforms under European sovereignty constraints. That post treated “Europe” as broadly singular. The reality is …
Building AI Platforms Under European Sovereignty
The Problem Is Not Capability# Most conversations about European AI start with benchmarks: parameter counts, the gap between Mistral and GPT-4, how far behind …
There Are No Emotions in Runtime Software
I read a piece this morning arguing that you shouldn’t trust software you didn’t “suffer” for. The premise being that personal struggle …
Streaming LLM Responses: A Partial JSON Parser for Structured Output
I’ve been building conversational AI features recently, and I hit an annoying problem: I want structured responses from the LLM (JSON with specific …
The Conversational Consent Pattern: Privacy That Actually Respects Your Users
I’ve been thinking a lot about privacy consent lately. Not the legal side (though that matters) but the experience side. The way we currently handle …
Seven Dollars and Twenty-Five Years: Building an Expo Plugin with AI
I’m building an AI training tool for small language models that ultimately deploys on mobile. Nothing revolutionary there. It’s a natural endpoint …
Building 3D Scenes Through Conversation: An MCP Server for Geographic Objects
I’ve been noodling around with the Model Context Protocol lately, and I wanted to share something I’ve built that demonstrates what I think is a …
Rendering Conversational 3D Scenes with Three.js and MCP
In my previous post, I promised that the scene data was portable. That you could feed it to “a game engine, a GIS tool, a mapping API, whatever.” …