Field notes from the practice. Architectural thinking.
Observations, experiments, and notes on AI governance, architecture, and the practical realities of building this work into organisations. Written by Navitec, for the practitioners and leaders we work with.
The AI Already Switched On
A field note on what a Copilot and agents check found in a small-business Microsoft 365 tenancy, and the settings that put it under control.
How a Language Model Is Actually Built
A practical guide to training a language model from scratch, turning it into a chatbot, and why the only real limit is compute.
Testing Basque First on a Single Box
A field note on training nine small language models from scratch to see whether the order a model learns its languages in leaves a lasting mark. Mostly it does not, and why is the interesting part.
The Shadow AI You Cannot See
A field note on a shadow-AI detector that runs entirely on your own machine, and what it says about governing the AI nobody authorised.
The Shape of UK Public Spending
A companion to the Tax Policy Associates map of the 85 UK taxes, built from the same OBR data. An interactive map of every pound of public spending in 2024-25, arranged by what it buys.
Retrieval Without Infrastructure
A field note on browser-resident AI, after a weekend spent putting Chrome's local model to work.
Design Authority at the Speed of Decisions
A perspective on what changes when architecture review moves from quarterly to per-commit, written after building a working multi-agent design authority panel.