MG12 design

MG12 design · est. on 30 years of delivery

Product & platform leadership
for the ML era.

I design and ship ML-heavy, scalable systems, and lead the teams that build them. CPO for hire, ML consultancy, and platform architecture for companies that need innovation they can actually stand behind.

30+ yrs
transforming clients’ businesses
100%
client retention, 20+ years
5:1+
typical delivery ROI ratio
90%+
customer journeys migrated to self-serve

New, but proven. Innovation with heritage.


Writing

How I think about building things that measure themselves, and about the data that has to be right before any of it works.

One conversation, two engines

Whole journeys are being built into chat models, which cannot guarantee the same result twice. The fix is not a form: it is one conversation with a deterministic process behind half of it.

Read the article →

The requirements nobody writes down

Requirements now become a live system in days, which makes the half nobody writes down decisive. Usability, accessibility, performance and scale are architectural, and cannot be added later at a sensible price.

Read the article →

The router is the architecture

Splitting by intent beats one engine for every query, but not because specialists are cleverer. And the router then becomes the most important component you have, running on the least information in the building.

Read the article →

Your threshold is a function of the question

Retrieval gates get tuned against test questions written by people who already know the answer. Real customers ask worse ones, the same correct document scores lower, and the gate throws it away.

Read the article →

The whisper game

Coding agents decompose a multi-step job into a chain of functions passing data along. By the end of the chain the data is a rumour of what arrived. Orchestration and a canonical entity are the fix.

Read the article →

One truth, delivered to the edge

A single customer view that lives in a warehouse is a reporting asset. The same view, present at every decision point, is a different product; and for ML, context is the largest lever on quality.

Read the article →

Circles, not lines

You can engineer an aircraft that flies and lands, and still build the wrong aircraft. What separates a working set of functions from a service worth running is a control loop.

Read the article →

All articles →

Have a problem worth solving well?

Tell me what you’re trying to build, fix or scale. I’ll tell you straight whether and how I can help.

Get in touch