AI Engineer
Eight years turning unstructured chaos into systems that think — lakehouses at 145-source scale, agents on Bedrock, and products people actually open.
“Most data work dies in a dashboard nobody opens. I build the boring parts so well that the interesting parts become possible.”
Who is doing the work, and why it tends to hold up.
Eight years ago I started as the person who fixed other people’s pipelines. Today I lead the design team on an enterprise ingestion platform pulling from 145 source systems into a three-zone lakehouse — and I still fix the pipelines, because that is where the truth lives.
The unusual part is the second half. I ship interfaces, motion and brand work with the same rigour I bring to a Spark job. Macrova, my nutrition product, went from sketch to store on both platforms. The engineering makes the creative work survive contact with reality; the creative work makes the engineering worth looking at.
Eight years, three employers, one throughline.
Leading design and architecture for an enterprise ingestion platform spanning 145+ source systems.
Four years inside the machine, on the hardest distributed data problems enterprise customers had.
Global retail engagement: moving finance data off Teradata and onto Google Cloud without breaking the month-end close.
Top of the programme — and shipping a custom Android OS to strangers on the internet at the same time.
The stack, set as a specimen sheet. Pick a face.
14 of 14 shown · bars indicate depth of use
Cloud
Four years inside the machine
Cloud Engineer, Big Data at Amazon Web Services (2021–2025). Worked directly with enterprise customers on the hardest distributed data problems, and was recognised internally as a Subject Matter Expert for AWS Glue.
Three pieces of work, photographed and captioned.
Plate I2024 —
Founder / Product Engineer
An AI nutrition system that looks at your plate and does the maths. Point the camera at food and get recognition, macro estimation and a plan that adapts — shipped on Android and iOS.
Plate II2025 —
Lead Architect · Liberty Mutual
An enterprise ingestion platform reading from 145+ source systems into a three-zone lakehouse, with an Audit-Balance-Control framework that governs every single run.
Plate III2023 —
Creative Development
Generative pipelines put to work: prompt-to-sequence video for brand film, Bedrock agents wired into governed enterprise data, and cinematic scroll-driven interfaces.
Numbers that survive being checked.
since 2018
one platform
enterprise + personal
cloud · data · ai
to real users
aws · gcp · azure
Five, across the three major clouds. Renewed rather than collected.
How a sentence becomes a system that runs at 3am.
01
It starts as a sentence, not a ticket. What should exist that does not? If I cannot say it in one line, it is not ready to build.
02
Sources, volumes, constraints, prior art. Understand the domain before touching a keyboard — most bad architecture is a research failure wearing a technical costume.
03
Zones, contracts, failure modes. Draw the system until the code becomes obvious. The diagram is the deliverable that saves the most money.
04
Python, Spark, SQL, Terraform. Build it once, build it reusable, build it observable. A pipeline you cannot see inside is a pipeline you do not own.
05
Orchestration, CDC, audit-balance-control. If a human repeats it, it becomes a workflow. If a workflow fails silently, it becomes an alert.
06
CI/CD, infrastructure as code, staged rollout. Shipping should be a routine with no adrenaline in it.
07
Cost, latency, skew, spend. The system gets faster and cheaper after it goes live — that is the part most teams never schedule.
How this was made, and how to reach me.
Let’s build something that outlives the demo.