What Technical Diving Taught Me About Building AI Models
At sixty metres there is no undo button. Every technical dive starts long before I enter the water, with a plan for gas, depth, time and every way the dive could go wrong. I have come to believe that the same mindset is what separates AI models that work in the real world from models that only look good in a demo.
Plan the dive, dive the plan
Technical divers write the whole dive down before they get wet. Depths, stop times, gas switches and turn points are all decided on land, where thinking is easy. Underwater, the job is to follow the plan and notice the moment something drifts from it.
I build AI models the same way. Before any training starts I write down what the model must do, what good looks like, which numbers I will measure and when I will stop. That plan keeps the work honest when results start to look exciting.
Checklists are not optional
On a closed circuit rebreather, the machine keeps you alive by managing the gas you breathe. Rebreather divers run a long checklist before every single dive, even after hundreds of dives, because the one time you skip it is the time it matters.
My data work has its own checklist. Where did this data come from. Are there gaps or duplicates. Do the totals match the source. Does the result still make sense when I look at it from another angle. Nothing goes forward until every box is ticked.
Redundancy keeps you calm
Divers carry backups for anything that could fail, from lights to gas supplies. Having a second option is what allows you to stay calm when something goes wrong.
In AI work that means keeping a simple baseline model next to the clever one, saving every version, and making sure a person can always check the output. When something breaks, there is a calm path back.
Calm thinking under pressure
The most important skill underwater is not strength. It is staying calm enough to think clearly when the plan changes. The same is true when a dashboard shows a strange number the night before a big meeting, or a model behaves differently with real users.
Stop, breathe, think, then act. It works at depth and it works at a desk.
Precision is a habit
People sometimes ask why I describe my AI work as surgical. It is because the habits I learned in the water do not switch off when I come back to shore. Careful planning, honest checks and calm decisions are what turn data into something people can trust.