Notes on the gap between research and production.

Large language models are extraordinary at language and reasoning over text. They are not optimisation engines, they are not planners, and they hallucinate with total confidence. Here is why none of that fits the core of StowAI, StackAI, and JobAI, and where an LLM might earn its place later.

We built our RL systems to work with terminal operators, not around them, because the hardest problems in logistics need both computational power and human judgment.

Getting clean, reliable data out of terminal operating systems, or any industrial environment, is harder than any model you will build on top of it. The gap between what is logged and what actually happened is where most projects quietly fail.

Deploying optimisation in container terminals means working within constraints that have nothing to do with mathematics. It is about earning the right to change how people work.

We talk about RL as though it optimises things. But the real insight is subtler. RL is a framework for making decisions, and optimisation is just a by-product of making good ones consistently.

A practitioner's honest take on deploying RL in one of the most operationally complex environments in global logistics, and knowing when to reach for a different tool.