The German medical system has often been criticized for being only reactive rather than proactive, and too focused on acute symptoms rather than looking at the body holistically. Those problems are exploited by a huge alternative pharma sector, using calm, empathic health practitioners to simulate a holistic analysis of the body and sell placebo drugs at insane markups.
Modern language models and their increasing capability to interpret visual data allow for actual holistic analysis of health issues and personalized treatment proposals at virtually no cost. The problem is just that your aunt won’t dump all of her health documents into Claude and provide instructions on how to interpret them.
That’s why Alex, Janik, Maxim, and I set out to build Medifold, a platform that aggregates your health related data and allows for analysis via large language models. The functionality is super simple, you have a single upload button for any health related document. Once something is uploaded, we automatically classify it and store it within its respective category. Then, with a single mouse click, you can generate a holistic analysis of your current health and get individual tips on how to fix it.
So far we support the following data:
- Diet-related documents (images of individual meals, images of grocery shopping receipts, PDF receipts)
- Journal notes you can directly write on our website
- Selfies and photos of your body that show health indicators
- Medical documents (blood work results, doctor’s appointment reports, etc.)
In the future, it would make sense to onboard a lot more data sources to get even better analysis. The obvious first choice is fitness and workout data from wearables and trackers like Apple Health data, Garmin, Strava, Fitbit, Oura. We wanted to already include that during the hackathon, but unfortunately getting access to that data takes verified developer accounts that we could not get over the weekend.
Medifold generates:
- A brief health analysis with focus on pain points
- Diet tips
- Fitness / exercise tips
- General lifestyle tips
- Supplement proposals
- A grocery shopping list based on previous preferences combined with current diet tips and supplement proposals
We were astounded by how good it actually works. I’d had some back problems the week before the hackathon and included my doctor’s report of symptoms in my test profile. Literally everything my doctor proposed was included in the tips!
Medifold is live for you to try it out here! ( In contrast to other hackathon projects, we even spent some time on data security for this one.) The prototype was built in ~1.5 days during the TUM.ai Makeathon. Thanks for organizing a great hackathon guys, and huge props to all winners!