The week I plugged my AI into my health data
The Links
Making a project data room will help give your AI better context and help reduce hallucinations.
A source inventory, conflict log, missing data inventory, and duplicate checking will help your AI write more reliably. I think this could have an interesting combo with a Karpathy Wiki. Pointing your AI to this data room will keep its writing more factual.
Love the idea of changing these hot softwares to make them your own.
- You understand them.
- You update what's there to make it your own.
Now I want to make my own version of Hermes.
Human consumption of the Internet is emotive, the AI consumption of the Internet is factual. Websites will need to do both. I listened to this while running intervals. I'm struck with a desire to buy a high end sound system.
It's not about using AI to make more crap for humans. It's about hooking up AI so it can read, understand, and act on what's there. We don't need another dashboard. We need our AI to act on the issues the dashboard is highlighting.
William Hahn from FAU's MPCR Lab recommended.
Play on 1.75x speed. Key takeaway for me: the way you did it and were successful won't be the way others do it to be successful.
Awesome, But Not AI Related
Jezz has gotten old. So have I. Loved discovering they've got a new series.
More rhymes than lines in your database. I listened to this on repeat while stuck on the tarmac waiting for, God knows what. Love this track.
Helps me understand my son more. I feel better for the future of humanity after seeing this. 🙄💀
Something I Built This Week
Gave my AI access to my health trackers.
Five trackers, five apps, five views I don't love. None of them give me what I actually want, which is to sit down with something that's seen all of it and have a conversation. How am I doing. What should I focus on. Pull me up a graph of my shitty sleep score from all the sources and tell me to take magnesium. With every number I generate, in one place.
Cost: 90 minutes for four of the five. That's the wild part.
What worked: Garmin, Oura, Strava, Whoop. I have very little idea what OAuth and API tokens actually are. I just did what Claude told me, and it worked.
What didn't: Hume. It's a scale that scans your body like a poor man's DEXA scan: muscle composition, hydration, fat percentage. Great product, lovely company, no API.
The kludge: I send Hume a PIPEDA (Personal Information Protection and Electronic Documents Act, Canada's privacy law) request on a weekly cron (scheduled task). Hume was super nice about it. They sent me the data and said they were happy to resend any time I wanted it.
The starter prompt:
I want my AI to have daily access to my health tracker data, Garmin, Oura, Strava, and Whoop. Help me get the four of them connected so the data lands somewhere my AI can read it every morning. Start with the one that's likely easiest and walk me through end to end.
This is a link to my training dashboard, it's based on the data I pull through this build.