I regularly get health tests like blood reports, BMI checks, body composition checks, prescriptions, etc. The biggest issue I face is tracking and comparing the data over time. I have tried multiple ways, like using Excel or tracking in my journaling apps or even using monsieur GPT to keep track. It’s all a pain.
Requirements
- I want a simple way to upload my reports, get the metrics transcribed, and then chart the data over time.
- I would like to keep my data private. All big model providers seem to be bleeding VC money and I feel ads are a simple revenue stream that all the providers can tap in. I already suffer with millions of spam calls and emails per day and I would NOT like to add to that.

- I would however like to use the structured data to talk to an LLM and get insights, suggestions, etc. A good approach would be how Hammer AI or DuckDuckGo is doing it. They give an interface to talk to an AI model and anonymize the conversation. Maybe a similar approach would work here.
- The platform/app should be open source.
Rough Solution Highlight
- Build a simple React dashboard with a Node server. This will be the entry point for the user to upload their reports and interact with the data.
- Pack the whole project into a docker executable. User can choose to self host or use a freemium cloud version.
- No PII taken from user. Just simple username pass with 2FA. Email optional for notifs etc.
- User can choose to either connect their AI model key or use a freemium (only for cloud versions) provided model. (Maybe like x free reports per y duration).
- The server will use a relatively small AI model to transcribe the data into a structured format. Might need some prompting.
- The structured data will be stored either completely on user device or in an encrypted format on a DB. The decryption key will be on user’s device and not backed up anywhere.
- No automated syncs? Maybe user can choose if and when to sync data b/w devices.
- Make a reports view and map the transcribed data to the location in the PDF. Like highlighting the source block.
- Build a metrics dashboard. Figure out how to show details that are in concern or under active notice of the user.
- Build a chat experience and give a tool call to the LLM to get and process the data. There are a bunch of open source chat experiences like Libre Chat which can be used here.
- For future, would be cool to let users share the information to a doctor for history or a semantic search for the information. For eg I search for my weight metric and the system gives me the data and also maps it. Or maybe I search a bunch of metrics and it generates a small PDF/UI report.
Conclusion
This is definitely a WIP and would need many iterations. But I guess it would be fun to build an MVP. So much to explore!

Leave a Reply