Exploring giving a genetic genius (o3 pro Deep Research) lots of context (workout logs, personal notes, a PDF of a canonical swimming textbook) to build a custom trainer.

Love this idea! I wonder how this could one day come to incorporate the true value of a coach on deck, which is to see the stroke and offer guidance. Some self-coached or remotely-coached swimmers film themselves underwater, which your system might reasonably consume at some point. I have worked with swim coaches in several formats, and the coach on deck providing instant feedback has been by far the most useful to me. Perhaps this is why one of the most common formats for swimming is the team practice, since one coach working in real time can watch a dozen swimmers and give feedback.

I had the pleasure of learning from Terry Laughlin at a Total Immersion camp in 2014. (I learned to swim as an adult for triathlon.) Terry loved the water so much that we could earn ourselves an extra minute of rest between sets just by asking him to show us a skill again. Far and away my most helpful and memorable swim instruction.

I am _very_ impressed by what the tech is capable of doing; slurping up pdfs, interpreting screenshots, and coming up with a plan.

But having gone the same route and relying on the book + Terry Laughlin's videos, I don't yet see how different this is from not using AI at all. If I'm feeling breathless, I look up the book for drills. If my shoulder aches, I look up the book for specific drills. I am very skeptical that slurping the pdf into AI would synthesize a new insight. At best it can be an excellent semantic index.

I think the OP does themselves a great disservice by not seeking a qualified swim coach to start with. Swimming is a deeply experiential workout and technique is very important. You can learn a lot of bad habits by swimming badly; 50m freestyle in ~2mins is, as example, showing that their technique is very bad right now (I suspect most swimmers here can probably kick 50m in <2mins with just a kickboard).