I'm Matthew Reda. I sit with marketing and finance leaders, figure out what's actually being decided, and build the smallest Bayesian system that can answer it. Sometimes that's a 60-line model. Sometimes it's a year-long platform. Always honest about what we know and what we don't.
The work doesn't start with a model — it starts with a half-hour conversation about what the business is actually trying to decide. Everything below is downstream of that. Six commitments, and one rule: if the answer's boring, that's still the answer.
Most measurement work fails before any model is fit. A team commissions a study to answer a question nobody's actually going to act on, or builds a dashboard for a decision that gets made on instinct anyway. The first thing I do on any engagement is sit with the people who own the budget, the P&L, and the brief — and figure out which decision actually moves if the answer comes back differently. Sometimes the right call is to not build the model at all. (The invoice is smaller; the relationship is better.)
Then I apply statistics carefully and honestly. Pre-specify the likelihood, validate the data, fit the model, check the diagnostics, report with HDIs. The training is from biophysics at MIT — single-molecule STORM imaging, where every pixel is noise until you treat it as a measurement problem. Same instinct, smaller microscopes: write the model down first, decide what would falsify it, then look at the data.
Honest reporting is the long-run business outcome. Stakeholders trust ranges that hold up. Finance trusts numbers that survive audit. Analysts trust code they can read. The compounding effect of telling clients what you actually know — and what you don't — is the biggest thing I have to offer. The Python tools, the agentic systems, the FastAPI services are how that compounding gets shipped. Also: I genuinely enjoy this. If you find a friend who lights up at the words "posterior predictive check," hire them.
{m.body}
Most of these started as a Sunday notebook and graduated. A few are still cooking. All open source — clone them, break them, file an issue.
{p.blurb}
The most recently updated public repositories, fetched the moment this page loads. If something looks half-finished, that's because it probably is.
MIT biophysics, then a half a decade in measurement science. The throughline: take uncertainty seriously, write the model down first, ship code, sleep well.
{e.blurb}
I take a small number of engagements at a time. Every one starts with the same question: what is the business actually trying to decide, and what would honest measurement of it look like? The deliverables are downstream of that. The conversation is free.
{e.body}
The best engagements start with a question, not a model. Send a short note about the decision on the table — the budget shift, the threshold, the trade-off — and what's currently being used to support it. I'll tell you whether honest measurement can move the needle, or whether you should save the budget and buy your team lunch instead.