Prodity: Product by Design

There are genuine substantive points buried in this episode - the population-specificity problem in ML model transfer, the paywalling of health system data, and comparing AI hallucination risk to existing human scribe error rates. However, the insight-to-filler ratio is poor: roughly half the runtime is consumed by triathlon discussion, baking company backstory, energy management book promotion, and product recommendations that have zero relevance to a B2B operator.

“the AI algorithm that was trained and learned on one population, you need to test it in your local population to make sure it works”

“by firewalling, paywalling that data such that I can't use it to train an algorithm becomes challenging”

The framing that AI hallucination isn't categorically more dangerous than existing human error (scribes make mistakes too, processes already exist) is a moderately fresh and useful reframe. The 'Gen AI is still underhyped' contrarian claim is stated but never substantively argued. Most other points - clear purpose, support network, market differentiation - are well-worn entrepreneurship platitudes.

“I am fascinated by the idea that we need a new ethics statement for AI”

“when they compare AI-generated patient notes to those written by physicians, the computer is generally seen as more empathetic than the physician”

RJ is a genuine practitioner - 22 years running a healthcare software consultancy, a congressional grant for military diabetes management, and 20+ years on a specific retinopathy detection project. That is real operational depth. However, he runs a small professional services firm, not a scaled product company, and a significant portion of the conversation drifts toward lifestyle content that dilutes his domain credibility.

“we got our start with a congressional grant for the development of a diabetes disease management solution of military healthcare. So we were working with like Walter Reed Army Medical Center and the Boston VA”

“one of the projects we've been involved with for 20 plus years now, is the detection of diabetic retinopathy”

A handful of concrete anchors exist - Walter Reed and Boston VA by name, $20 vs $200 OpenAI subscription tiers, $100K-$120K developer salary as a benchmark - but the most substantive claims are left vague: paywalled data is described only as 'cost prohibitive' without any figures, health system data vendors go unnamed, and statistics like '90% or whatever' underscore a pattern of imprecise assertion.

“you need significant backing to then get access to that data to be able to have a variety of information”

“pay for the $20 subscription of... to get the better service kind of thing. And now there's like a $200 level to get the reasoning model”

The host asks broad, open-ended questions and never challenges a single claim, including the unsupported assertion that Gen AI is 'still underhyped' and that agentic AI is overhyped. Extended off-topic tangents on triathlon, baking, and the energy management book are allowed to run without redirection, and the closing questions (favourite product, favourite book) add no substance for a B2B audience.

“is there a part of the triathlon that you like the most and the least?”

“what advice would you give to somebody who wanted to start a company or is looking to get into this space?”