Have you or someone you love ever had a medical procedure that requires extensive recovery afterwards? You're typically briefed by your doctor on the overall dos and don'ts. For example, if you're recovering from surgery following a broken ankle, it might be: "No showers or tubs for 72 hours" or "After a few days, start wiggling your toes," etc. And when you leave the healthcare facility, you're handed a pamphlet that offers the details and timings for these dos and don'ts. These pamphlets vary by provider and facility and are often hard to follow.
I'd recently had a medical procedure myself when I met Jay and his son Grant at the CodyMD hackathon in downtown Portland. I was becoming more adept at agentic engineering and wanted to see if I could help another team build their product. That's how I ended up meeting Jay. His pitch was simple: "Why do we have all these pamphlets? Shouldn't we have an app that can help you track your dos and don'ts?" Instantly I thought yes, 0f course we should. Simply put, that’s a much better standard of care.
We decided to design and build an app that does everything the paper pamphlet does but makes it simpler to follow, focusing on what the patient needs to do today, and notifies and celebrates when restrictions are lifted. That first shower after a few days is certainly cause for celebration, right?
But there was a catch... How were we going to take those instructions in the pamphlet, reorganize them, and get that data into our system? We saw a few options:
- We ask the providers to manually recreate the instructions using a content management system we provide.
- We use AI to translate the instructions and ask the patient to photograph their instructions and upload them.
- We upload the instructions for the providers and use AI to translate them, then ask the provider to approve them.
We knew the answer was likely #3, but even though providers would approve the We knew the answer was likely #3, and even though providers would approve the procedure instructions used in Post Pal, we wanted to introduce some certainty into translating the instructions. First, we developed a system that read through the PDFs without AI. However, while there's no risk of hallucination using an OCR tool and you’re working with a set of rules, the outcome was nonetheless poor. The translations were rough. Once we hooked Claude AI into our system, we were cooking with gas. Its ability to make decisions while reading through the PDFs was key. And since we wanted to reduce the chance of hallucination, we also hooked up Mistral's OCR service and introduced redundancy and a scoring system into our translations with perfect results.
This is why I say the Post Pal app is AI-enabled, but not AI-powered. Without AI, the cost to translate or recreate those pamphlets would likely be too high for a payer or provider to say yes. But with AI, we can take a provider's pamphlets and recreate them in the app with a high degree of confidence.
Spoiler alert: we didn't win the hackathon. We were solving what appeared to be a smaller problem than some of our competitors, but it was clear our app would actually solve the problem. The challenge will be to build it out, beta test it, and find payers or providers to pay for using Post Pal.
Working with Jay, I've built out the web app and mobile apps for Post Plus using Next.js and React Native, with Next.js providing the API, web app, and management system. We're currently looking for a provider to beta test our app with. While we like how our app solves the problem, we look forward to the next phase of getting feedback from actual patients.