Stories

The systems I build and run, and the experiments that got out of the lab. Different scales, same standard.

From practice
Architecture

The operating system behind OsteosOnline

An osteopathy network doesn't run on a website. It runs on an operating system: knowledge, operations, agents, evidence, and a loop that makes itself smarter. One map, a whole company.

Read story
Migration

Not one medical site, but the factory behind it

OsteosOnline moved from WordPress to Next.js, but the real work was not a prettier site. It was the factory underneath: data, routes, redirects, structured data and tests.

Interactive
Experiments
Operations

From accountant screenshot to permanent tool

A WhatsApp screenshot of financial tasks. Thirty minutes later, three command-line tools that did not exist before. The difference between doing the work and building the thing that does the work forever.

Read story
Lessons
Architecture

Not the model, but the harness

Dependency versus sovereignty. Why I don't pick models, but build the harness that makes them swappable. Exit power over benchmarks.

Read story
Architecture

From prompting to compiling

Intent goes in as a brief. Agents act as compiler passes. Eval gates act as the typechecker. The output only ships if it passes every check. I treat AI agents like a compiler, not a slot machine.

Read story
Around osteopathy

16 years as an osteopath is systems thinking for the human body. 10 years of technology and AI is that same thinking for software, and actually building it. The two meet in an AI-first ecosystem: a digital right hand for every owner, and three layers around osteopathy for practices, knowledge sharing and patient guidance. The common thread: the expert must not stay the bottleneck.

Live

Coel.ai

The knowledge layer that scales expertise

One knowledge base, assistants and standards, deployable anywhere. Distributing professional knowledge at scale for associations, schools and expert groups.

In progress

Lumi

The patient layer between consultations

Continuity of care beyond the treatment room: preparation, follow-up and education for parents and patients.

Something like this for your business?

We start with an introductory call. Then we decide together whether an analysis of your work is the right first step.

Book an introductory call

Under the hood

The operating system behind OsteosOnline

No slides. The full system map of a real business: what exists, where the truth lives, how the agents work, and how every loop makes the system smarter.

defines
start
drives
delivers
improves
OsteosOnlineHow the system works

One map: what exists, where the truth lives, how we look at it, how agents work, and how work turns into evidence and learning.

Knowledge

How we work

Canonical source: repo / markdown / data

  • RouteTemplates · areas & flows
  • RouteParts · steps & tasks
  • Runbooks · data & execution
  • Procedures / SOPs · explanation & standard
  • Agent Recipes · machine-runnable

Toolbox

Integrations
Google Search ConsoleGoogle Analytics 4Meta AdsGoogle AdsSendGridOdoo

Inflow

Signals
  • New starter · website / intake
  • Partner request · form / contact
  • Patient search · website / Google / maps
  • Booking made · schedule
  • Analytics signal · GSC / GA4
  • Content idea / question · team / market

Operations

What runs

Canonical source: Operation entry (repo → later Supabase)

OperationStarter onboarding · Luke
Progress3 / 6 steps
Statein_progress

Progress

  • ✓Process intake
  • ✓Create profile
  • ○Schedule intro
  • ○Add practice
  • ○Review & GO
OwnerLuke
Work fileGitHub issue
Next actionSchedule intro
BlockersNone
Done whenWrapped up & followed up

Agents

Execution

Work through an Operation Context Package

Operation Context Package

  • State · live status
  • Runbook · how to run it
  • Knowledge · rules & knowledge
  • Data · sources & context

How an agent works

  • 1. Prepare · get ready
  • 2. GO · human approval
  • 3. Action · external action
  • 4. Evidence · capture proof

Work file

Issue

GitHub issue per Operation

  • Queue
  • Discussion & context
  • Handoff
  • Review & GO
  • Agent runs (log)

Not a source of truth for state.

Evidence

Proof & artifacts

Canonical source: evidence refs + external systems

Transcripts · Drive
Contracts · Drive
Profile page · Website
Emails · SendGrid
Reports · Drive / PDF
Bookings · schedule

Analytics

What we measure
Signalraw data↓
Metricmeasurements↓
Reportoverviews↓
Insightinsights

Source data (today)

GSCGA4BookingsAds

The source data already exists, the model follows.

Learning

Back into the system
What worked?analysis↓
What can improve?insights↓
Procedure / SOPadjust↓
Runbook / Recipeimprove↓
RouteTemplateupdate

Learning feeds the Knowledge layer.

Principles

Always true
  • One source, many viewers
  • One truth per concept
  • Prepare → GO → Action
  • Working method first, automation second
The fundamental loop
Signals→RouteTemplate→Operation→Work file→Procedure→Agent Recipe→Agent run→Output→GO→Action→Evidence→Learning→Better knowledge↺

The system behind OsteosOnline. One business, one map, and the loop that makes it smarter.