Every project below is in production or has been. Each one names what it proves, what it
cost me to learn, and where the boundaries are, including the parts that did not work.
No concepts, no tutorials, no side projects that ran once and died.
In productionFlagship
AroFlo data platform
Architect · Executor · Orchestrator
The finance platform I designed, built and run. It pulls a job-management system,
accounting, HR and time-tracking into one governed warehouse that leadership's
reporting reads from, and, more importantly, it is built to catch its own mistakes.
Two independent pipelines load the same source into two databases so they can be
checked against each other.
Investigated, built and validated against production
One row per vehicle, plant item and piece of equipment: what it is, when it was last
safety-checked, by whom, and its latest odometer and hour readings. It joins an API to
a spreadsheet export because the readings are write-only and cannot be pulled from the
API at all. The extract it replaced had been reading one page of nine.
One extraction pattern, built for a clean supplier and proven against a much messier
one, a folder mixing invoices, credit notes and purchase orders together. Different
regulatory domain to finance, same discipline: the incremental version that shipped
first would have quietly overwritten three years of invoice history, caught by a test
written to force that exact failure before it ever touched production.
120→112PDFs classified, only the real invoices parsed
13→2rows the first version would have left standing
Power BI does not point at raw tables. It reads a typed SQL view layer with a
read-only login, so business logic lives in one place and a report cannot silently
invent its own definition of a number. Seventy tables and a hundred and ninety-four
measures in one model, with row-level security.
70tables in one governed model
194measures, row-level security
Write-up in progress
You are on it
This website
Static site, own deployment pipeline
Hand-written HTML and CSS with no framework, assembled from partials by a small build
script, deployed by pushing to a public repository that the host pulls automatically.
A private build repo holds the source; only what should be public ever reaches the
public one. Boring on purpose: it costs nothing to run and cannot break at 2am.
Write-up in progress
Get in touch
I am happy to walk through any of these builds in depth: the architecture, the decisions behind it, or the parts that failed and what I changed afterwards.
