Hey hey 
AI is everywhere these days, and we were talking about it on Slack yesterday. That got us thinking that it would be awesome to see what you have been building with Meteor + AI.
So, who has some cool projects and stories to tell us about this? I’d love to feature interesting work on our YouTube channel and blog.
I use AI for a couple of things.
At the moment, I send the first one directly to OpenAI to be handled via gpt-4.
The second one, I send to AWS Bedrock to be handled via anthropic.claude-3-haiku-20240307-v1:0
If anyone has recommendations for better, faster or cheaper models, please post them.
Around August of 2024 I started building an AI medical assistant that integrates directly with EHRs using standard protocols like HL7 and FHIR to retrieve and send data. It started simply as a transcription / scribe that was really good at recording audio and transcribing it. After that initial phase, I introduced the ability for it to reason and generate notes (SOAP, DAP, etc.) based on the transcribed data - saving physicians 2-3 hours on average each day documenting their visits. One physician summarized its capabilities as liberating.
The real-time nature of Meteor has allowed real-time insights to appear as soon as the data is available to share. Physicians are already burnt out enough as it is… technology should make their jobs easier, not harder, and thanks to Meteor, Ozwell has been a huge success. The time it took me to go from IDEA to MVP was like 3 days. In under a week we had physicians using it in patient visits. We had 30 minutes of downtime while moving to a new production stack (since it was literally just a dev system running meteor run and not a built system!) and we had physicians calling us telling us they can’t see patients because the system was down. They didn’t want to practice medicine without it!!
The code is 100% written in Meteor. We use Blaze, FlowRouter, FastRender, and many other popular packages to pull it off. I’ll be doing a more in-depth post in the future about the architecture behind it.
My “referral” link to try it is here (it tracks that I sent you). If you’d just like to try it, no referral or sign-up required, click here. The official website is https://ozwell.ai
Some screenshots:
Congratulations on this great app! How do you handle the HIPAA requirements when sending audio to an LLM for transcription?
Depends on the use case! Often times a BAA and “zero retention” policies can get you there. We own and operate our own datacenters with regular security audits and certifications like SOC2, HI-TRUST, etc. We currently store over 500 million medical records for some of the largest fortune 100 companies in the world – from 60+ different countries! We have a mixture of models that we run on-prem with our own GPU compute as well as a mixture of zero-retention / HIPAA compliant agreements with third-party services like OpenAI for various tasks.
Fantastic. And all this is supporting the app you described, or, are there additional services you provide as well?
Hey all - I’ve built (and posted) about this elsewhere too. I’ve built Slatebox as an AI text-to-visual app for building infographics, mindmaps, timelines, pitch decks, etc – then present them with one click. Here’s one I made for Slatebox with AI:
