Physician. I build open-source tools for health data. Santiago, Chile.

I work on what a dataset doesn't show you — and who goes missing in the gap.

Chile has no usable baseline for mental health. Without one, no public policy can be evaluated — budget, gaps, and priorities get decided on anecdote or on figures from a decade ago. The data exists; it's scattered across agencies, inconsistent between years, half of it locked in PDF.

obsm is the infrastructure that baseline needs: public sources consolidated, normalized, and traceable, published by commune and by month, with provenance recorded for every dataset and every run. It's not a clinical app. It doesn't triage, doesn't touch identifiable patients, and doesn't replace anyone's judgment. It exists so that arguments about mental health policy can be settled with numbers.

First series: communal suicide mortality, 2002–2023. Published by grouper and never by method — a safe-publication rule the project holds itself to, on top of statistical secrecy. One command reproduces it, and it won't publish at all if its reconciliation anchors don't hold.

Health systems ration expensive programs with a model that decides who counts as high risk — usually defined as who will spend the most, because spending is recorded for everyone. Someone in real distress who never seeks care spends nothing, so the model reads zero and calls them low risk. The need was there; the label couldn't hold it.

riskaudit is a tool for measuring that, and the tool is the point. Give it the scores a model produced and an independent measure of need, and it quantifies how much need the model leaves behind — weighted, with confidence intervals. It doesn't care what the model predicts or where: a hospital's readmission model, an insurer's cost model, a ministry's triage algorithm.

MD, Pontificia Universidad Católica de Chile · MSc Data Science, UC Chile (in progress) [email protected] · LinkedIn