Anecdotally, I have heard that hypochondria is a problem in this space. So I would want to know the ratio of incorrect self-diagnosis to confirmed diagnosis, and what percentage of the system's totally human and technical resources are spent recognizing false positives.
I think that this sort of effort is probably a positive one. Patients should have access to their records are are more likely to be able to centralize data for analysis by a 3rd party - with all the attendant risks - more effectively than any other party. Secure systems without conflicts of interest - ie, insurance companies - to aid multidisciplinary diagnoses over longer time frames is a good thing.
But the risk of AI hallucinations and doctor/diagnosis shopping are real. This post was light on details on the diagnosis, how many rheumatologists they saw to get it, what tests confirmed it, and if the treatment has been effective in a way that cannot be skewed (https://pubmed.ncbi.nlm.nih.gov/30139289/)