I've been tracking my own health data for quite some time now, blood tests, sleep, how I feel, diet. The problem was always that everything lived in separate silos inside a basic CMS I have built. Some time ago I started taking a master's degree, in nature science and I started studying statistics and while looking for some project to exercise and apply what I was learning I decided to resurrect my health cms project try to tie some basic analysis in it, but then rabbit hole digging began ahah. I began reading papers on biomarker correlations, EWMA smoothing, allostatic load. At some point I thought: what if I actually modeled this properly?

So I refactored my whole app into something a bit more modern, thanks to AI helping me during this process and now I integrated some of the most interesting findings I found in the papers (linked in the readme of the repo). I named it Synaps, because it's about connections and integrated system, and so far this is what it can do: - Extracts biomarkers from lab report PDFs/photos via Claude vision - Tracks sleep, diet, activity, wellbeing, body composition, medications, environment (Open-Meteo, I'm mad about this stuff) - Computes continuous-time EWMA, z-scores, trend detection, and within-person Pearson correlations across domains with lag 0–2 days, autocorrelation correction, and Benjamini-Hochberg FDR - Builds a typed knowledge graph (biomarkers, conditions, lifestyle, environment) with AI-inferred edges validated against the statistical correlations - Estimates allostatic load from a 17-cohort IPD meta-analysis consensus

Everything runs on your own hardware. Install is a single curl | bash. No cloud, no subscription, you can also use it without AI and just analyse all your reports manually, but you lose some important feature.

In the next future I would like to also test how it perform with local models using ollama, but this is what I have been using for my self and I like it