PEP Health turns public patient reviews into insight for hospitals and health systems. When I joined, its data platform ran on AWS for about $26k a month and was hard to change. Every new client or data source meant work nobody wanted to take on.
I argued early on that the platform needed rebuilding. The rebuild kept losing out to more urgent work, so I kept making the case, using the running costs, the failures and the revenue we couldn't onboard, until it became the priority.
Partway through the build, a security incident meant we had to shut the old platform down before the new one was ready. In the same period the company had serious revenue problems, and I was asked to cut my team of six (four data engineers and two web engineers) down to one.
I wanted to do that properly. One engineer was already working their notice. A junior who wanted to move on for their own development kept the time to find the right next role. For a mid-level engineer on probation, I negotiated a three-month fixed-term contract so he could leave with a job to go to. Then I brought in a strong contractor and later persuaded him to go permanent, because a company with limited revenue needed one person who could own a whole system.
With a team of two, we finished the build. We leaned heavily on AI coding tools to make up the difference.
Alongside the platform, I run the data side of commercial decisions. I choose and buy the platforms we collect data through, advise on what we can sell from what we hold, and train people across the business: SQL for the data scientists, self-serve dashboards for the executives, data modelling for the engineers.
AI wrote most of the new warehouse. It got us to a working version fast, but I'm not convinced it got us to a reliable one any sooner. Since then we've spent months proving what it built and rewriting the parts where it went off-book. Next time, people would write the core models. I wrote about it here.
I was promoted to Head of Data at PEP Health in February 2026. I've line-managed up to six people across data and web engineering, reporting to the CTO. That meant running 1:1s and development plans, hiring, managing performance, and the restructure described above. I've built the team across disciplines: data engineering, web engineering and now machine learning, and I've hired three data engineers into it. After the restructure I'm rebuilding it with a Senior ML Engineer, a role that will also help develop the data science team.
The work I'm proudest of rarely shows up in a metric. One engineer I managed was technically strong but kept waiting for someone to check his answers. The gap wasn't skill. He needed to see why the modelling mattered and to trust his own judgement, so I gave him the decisions, asked questions instead of giving answers, and made sure his wins were noticed. More on that here.
I also train people outside my own team. I ran GDPR training workshops at U Holdings when the regulation came in, and AI fundamentals training across the IT department at Engie. At PEP I've run SQL upskilling for data scientists, self-serve QuickSight for the leadership team and data modelling workshops for engineers, and I onboard people from other parts of the business onto the data.
I was hired to take over a platform that contractors had built and nobody in-house could maintain. I designed a new ingestion framework on AWS and a Snowflake warehouse, brought the company's data into it, and set the engineering standards around it.
Fixing the platform was half the job. The other half was making sure the company wasn't dependent on one outside party again, so I hired an engineer and got them up to speed, and the business ended up with in-house people who understood what it had.
Engie's AWS and dbt estate had grown faster than anyone had designed it. Costs were high, failures were increasing, and no one had a clear view of why. I reviewed it piece by piece, found the expensive bottlenecks, and re-orchestrated the codebase so jobs ran in parallel and failed less.
I walked the team through the new shape as I went, so they could keep it running rather than inherit another black box. I also mentored engineers on AWS and Python.
I built a multi-tenant BI platform from scratch, consolidating more than 300 client databases into one Redshift warehouse with QuickSight on top. The hard decisions were where to standardise and where to let clients differ, so that client number 301 didn't cost more to support than client number 30.
It's where I learned that scale is mostly a design problem, and the time to solve it is before you need to.
Want this kind of work done for you? Work with me →