ML pipelines that don't deploy. Builds that take forever. Tests that nobody trusts -- or don't exist at all. I've fixed these at European unicorn scale. I'll fix them for you.
Let's talkI build self-service model lifecycle frameworks. At Bolt, while I spent 3 months on infrastructure work, the DS team created 100 new ML projects largely without data engineering support. The ability to experiment and iterate without waiting for DE was a force multiplier for productivity and culture.
I've cut build times from 40 minutes to 5, optimised Docker images, fixed flaky CI pipelines, and built autoscaling build clusters. Whether it's ML training jobs, C++ compilation, or Gradle builds -- slow builds are a solvable problem.
I systematically audit and optimise compute, storage, and processing costs. RAM reduced by 90%. Cloud spend per workload cut 30-90%. Most companies are massively over-provisioned and don't know it.
You grew on MVPs and duct tape. It worked. Now you have 50 models, 3 CI systems, and tests nobody trusts. I've been through the 10-to-1,000 journey. I know which foundations to lay before the next 10x, and which shortcuts to keep.
Automated training pipelines, self-service deployment, reliable CI, proper test suites. Retraining goes from quarterly manual babysitting to daily automated runs. Engineers get back to building.
How your team builds, tests, deploys, and operates -- assessed against what works at scale. Whether you have too much tooling or none at all, you get a prioritised roadmap of what to fix first.
Pipeline and build system design, build vs buy analysis, implementation plan. Get the architecture right before committing.
Quick-win cost reductions in compute, storage, and builds. The 30-90% savings usually have equivalents in your setup. ROI within the first month.
Embedded alongside your team: pipeline design, build tooling, CI/CD, cost optimisation, and knowledge transfer. I build the foundations, then your team owns them.
I'm Mark Cowan, founder of Potamus OU, based in Estonia.
For nearly 5 years at Bolt (one of Europe's largest tech companies), I built the model lifecycle platform that scaled the data science operation from a handful of people manually deploying a few models to 50+ data scientists independently managing 200+ ML projects. I also built the CI infrastructure, optimised build pipelines, and made engineering teams more efficient across the stack.
Before Bolt, I built satellite flight software at Open Cosmos (code that flew in space) and led teams of up to 11 engineers. As CTO at a defence tech startup, I architected flight software running real-time computer vision on hardware 80% cheaper than competitors, supporting multiple hardware platforms. I'm an Entrepreneur First alumnus (London, Cohort 5) and hold a patent for satellite payload development.
I focus on making engineering operations work: the pipeline, the build, the deployment, the cost, the reliability. The heavy stuff that slows everything else down.
A 30-minute conversation costs nothing. I'll tell you honestly whether I can help.
Or email directly: mark@potamus.ee