What changed

Governance systems in daily operation

11

from zero

Delivery roles documented with RACI

7

200+ responsibilities

Direct reports in biweekly coaching

8

sustained seven months

Presenters developed in the team

7 of 9

from one voice

Growth sessions, zero cancellations

13+

May–August 2026

Two senior hires to production

Day 3

onboarding rebuilt from zero

Where it started

Context matters more than a list of wins. This was the state of the engineering organisation in February 2026, and what stands in its place now.

Onboarding

No process; new hires spent weeks orienting themselves. Now a structured path with a checklist and milestone tracking, which put two senior hires into production on day three.

Code review

Informal — no SLA, no template, no ownership model. Now a documented process with a 48-hour first-review SLA, CODEOWNERS, rotating primary and backup reviewers, and quality gates in CI.

Training

Ad-hoc and self-driven. Now a catalogue with a lifecycle and a real budget attached: 200 hours a year, half of it carried over, tracked per person.

Technical decisions

Verbal, undocumented, unrevisitable. Now architecture decision records in Nygard's format, a quarterly Tech Radar in ThoughtWorks', and C4 models in Brown's — kept inside the codebase so they rot at the same rate as the code.

Sprint planning

Flat task lists. Now epics, user stories and sizing, against a capacity model that shows the supply side rather than only the demand.

Incident management

Non-existent. Now a four-level severity matrix, five defined phases, a rotating incident commander and blameless post-mortems, mapped to the relevant ISO 27001 controls.

Roles and accountability

Undefined. Now seven delivery roles documented with RACI boundaries and more than 200 numbered responsibilities — including, for each role, what it does not own.

Design consistency

A bottleneck on one person. Now a cross-functional guild with its own lead and a contribution model; roughly 80% of design decisions no longer reach the design lead. That last figure is an internal assessment, not an instrumented metric.

Direct delivery

The hardest work in the portfolio was architected and written, not supervised.

An ERP gateway on AWS

160+ commits on that one product over three months: a JWT authorizer refactored to RS256 with Parameter Store integration and property-based tests, async CSV ingestion on S3, Lambda and DynamoDB, a Storybook host built from zero with CDK and branch previews, an integration and end-to-end test framework, a Powertools logging migration across the whole codebase, 16 security vulnerabilities resolved, and ADR-021 through ADR-036 authored in-repo.

A multi-tenant SaaS for public administration

End-to-end architecture and tech lead. Event-driven on EventBridge and Step Functions, multi-tenant, defined in AWS CDK and deployed in eu-south-1, with NIS2 compliance designed in on day zero — region choice, encryption, audit logging and least privilege as architectural constraints rather than a retrofit.

A technology validation gate

Every technology entering production passed four steps: independent study, a proof of concept on a real use case, a Tech Radar placement backed by a written decision record, then a training session if it reached Trial or Adopt. No blind adoptions.

Team multiplier

A ritual that survives thirteen weeks, and a presenter base that grows from one voice to seven, are not attendance statistics. They are evidence that the system is no longer person-dependent — that it keeps running on the days I am not in the room.

• Eight direct reports in biweekly coaching 1:1s, sustained across seven months — run as coaching, not as status reporting.
• Thirteen consecutive Technical Growth sessions with zero cancellations between May and August 2026, presented by seven of the nine engineers rather than by one.
• A training system with money attached — 200 hours a year, half carried over, a catalogue with a lifecycle, and per-person tracking. Growth stops being a stated value the moment it has a budget line.

Structural value

Eleven governance systems in daily operation, from none. An organisation with documented decisions, defined accountability and active metrics carries measurably less operational risk than one running on folklore — and the documentation outlives whoever wrote it, which is the point of writing it down.

Metrics describe the system, never the person. There are no per-author dashboards, and there will not be: the moment a metric ranks people, it stops measuring the thing it was built to measure.

The pattern before this

This is the third time the shape has repeated: arrive where the function does not yet exist, build the foundation before the team, and leave a system that runs without me. Two earlier engagements are the evidence that F.technology is a pattern, not a one-off.

TokyDigital — an in-house engineering function from zero

Interim CTO for twenty-two months, hired to internalise an entirely outsourced technology function.

The mandate

Interim CTO, twenty-two months. The company was fully outsourced with no internal IT; the brief was to build the function from zero and internalise it — a founding mandate, not a caretaker one.

Foundation first, alone

Started 16 October 2023. A proof of concept was running inside the first month, and the platform foundation was complete by January 2024 — roughly three months building it near-solo before bringing anyone into a codebase that already existed.

Then the team

Two engineers hired in April 2024, beta in May, released in August, and presented publicly in Madrid in September 2024 — a WhatsApp Business service taken from proof of concept to public launch in eleven months. A fully outsourced function cannot ship a public SaaS in eleven months; the product is the proof the insourcing worked.

Cambiomarcia — a re-platform, not a refactor

A full re-architecture of the platform, from a monolith to an event-driven serverless system. The two topologies are the evidence — any engineer does the scalability arithmetic from the shapes themselves, so no latency or cost figures are needed. This is where the AWS and serverless depth actually formed.

Before

A Laravel monolith on two EC2 instances — one serving internet traffic, one running cron and queue jobs — with no load balancer on either.

After

A single-page front end; a GraphQL API on AppSync; DynamoDB; Cognito for auth and identity; and Step Functions with SQS for long-running async work.

The platform later passed Deloitte's technical due diligence: technology was one of three assessment areas, alongside the business plan and the market, and the architecture drew the analysts' attention. It is the one credential here that is externally validated rather than self-asserted.