I'm a diabetic (type one, since 1988) who has also been doing ML startups for the last 15 years. My HbA1C scores have always been below 6, controlled with a two to four blood test per day and long/short (currently Novorapid and Lantus) insulin regime. I've always been sporty, and have done some more difficult things like hiking to Everest base camp without a porter, cycling from Montreal to Lake Placid and back, and a few 24 hour rogaining events. In other words, broadly I've managed my diabetes quite well.

I've always been tempted to do something with ML and the datasets I collect off my own body, but I have never seen this point. In my experience, the thing that allows for good diabetic control is a doctor who can help a person understand their own body's reaction as a dynamic system, rather than a rigid schedule or an external system that tells them what to do. A product like this would be most useful in my opinion as a learning aid.

The place I see this kind of data being extremely useful is in the insurance industry. I currently pay a 400% premium on my life insurance due to being diabetic, but my control is extremely good. Were it to allow me to obtain reasonable insurance coverage, I'd be more likely to be willing to go through the data collection headaches involved with this kind of product.