Most people assume our most advanced AI work happened recently. It didn't.
One of the earliest AI projects we worked on with Dr Oetker is still one of the best examples I have of what makes AI actually work. It was cutting edge at the time. Honestly, in some ways, it still is.
What we did was jointly install high-resolution cameras across four stages of Dr. Oetker's production line, capturing every pizza as it was baked, sauced, topped with cheese and prepared for freezing. Helm Vision analysed approximately 50,000 pizzas every day, identifying overtopped, undertopped, and uneven products in real time. The result was a system that reduced waste, improved consistency, and gave the operations team live visibility into production line performance.
What made it work wasn't just the algorithm. It was the environment we built it in.
It was a closed-loop system that was controlled and self-contained. We weren't trying to stitch together five different legacy systems or wait on someone else's API roadmap. We owned the full stack, end to end, which meant we weren't dependent on a third-party vendor to ship a fix or unlock a feature before we could move.
That sounds like a small detail, but it’s actually quite significant.
