Data
Platforms
Lakehouse and warehouse estates on Databricks across Azure, AWS and GCP — alongside BigQuery, Snowflake, Synapse, and the Teradata and Hadoop estates they usually replace.
Data, Cloud & AI Leader
Technologies change.
Problems just get more interesting.
I lead by day.
I build by Sunday.
“Every AI programme is a data problem wearing a costume.”
In an AI-first world the leverage isn't the model. It's the sequence you run before you ever reach for one — and it's the same three moves whether the answer turns out to be a warehouse, an agent, or nothing at all.
Most failed platforms answer a question nobody needed answered. The cost of the entire programme is decided here, before a line of code exists.
Symptoms are cheap and plentiful. Causes are rare, and they are usually organisational long before they are technical.
Simpler architecture, measurable outcome. The good version is the one that survives handover to the people who have to run it.
Databricks, GCP, the model of the month — all replaceable. Everything else is tooling.
A career read the way you'd read a distributed trace. Width is tenure, nesting is scope, and the playhead runs the whole thing end to end. Drag it, or press RUN.
The dashed span is dataNX — running in parallel with every job since February 2019. Longer than any employer on this trace, and it has never stopped.
Enterprise architecture Monday to Friday. On Sunday, the experiments that keep the architecture honest. Neither one is the hobby — click either side to switch.
Teams and technology initiatives across data modernization, Databricks, GCP and AI platforms — complex problems turned into simpler architectures that survive the handover.
The Sunday Builds — real problems turned into working AI experiments on local and open models. No API keys, no vendor lock, no slides. Plus dataNX, teaching the same ground in public.
Three pillars, one job. The tooling underneath changes every couple of years — these are the places the decisions actually get made. Everything below that line is what the problem happened to need, picked up because it was the right answer at the time rather than the interesting one.
Lakehouse and warehouse estates on Databricks across Azure, AWS and GCP — alongside BigQuery, Snowflake, Synapse, and the Teradata and Hadoop estates they usually replace.
Turning tangled estates into systems that survive handover. Migration strategy, medallion design, governance at scale, and the cost model that keeps it affordable afterwards.
Model serving, retrieval and agent layers over governed data — and on Sundays, the same problems solved end to end on local and open models.
One problem, one weekend, one working thing. Local and open models — no API keys, no vendor lock, no slides. Plus dataNX, where the same ground gets taught in public.
Solving for what's next. If you're staring at an estate nobody wants to touch, a platform that outgrew its design, or an AI programme that needs an architecture behind it — that's the conversation I want.