automation · optimization · modeling · ai
systems that decide, adapt, and improve themselves.
matrices is the independent practice of sam cialek, built on fifteen years across quantitative finance, data science, and software engineering. Anyone can claim these skills. This page runs them: three small working systems, live below, no server, no libraries, one file.
live on this page
the demo is the pitch
Slide decks say trust me. Working systems say watch. Each panel below is a miniature of a real engagement, built with the same methods described further down, and everything it does is computed right here in your browser.
demo 01 · modeling
a posterior you can poke
Each press is one observation of a coin with unknown bias; the curve is the Beta posterior after your evidence, with the 95% credible interval shaded. The same machinery, scaled up, is how serif turns sparse biomarker draws into dose-aware recommendations.
demo 02 · optimization
descent on a curved valley
Gradient descent with momentum on a valley where the steepest direction rarely points at the answer. Real allocation problems have this shape: the objective is easy to state, and the geometry is where the work is.
demo 03 · automation
a pipeline that runs itself
Press run and a job moves through ingest, score, decide, act; confident cases complete unattended, and uncertain ones divert to human review. That is the shape of every agentic system worth shipping: autonomous by default, accountable by design.
what the practice does
four disciplines, one point of view
Decisions are the product, and uncertainty is part of the spec. Everything below is in service of systems that keep working after the engagement ends.
automation
Agentic pipelines, workflow orchestration, and LLM-powered systems that run unattended, with human-in-the-loop checkpoints placed exactly where judgment matters. The goal is not fewer people; it is fewer interruptions.
optimization
Portfolio and resource-allocation optimization, and decision analysis under uncertainty. Objectives are distribution-aware rather than expected-value-only, so tails and downside shape the answer instead of vanishing into an average.
modeling
Bayesian inference in Stan, PyMC, and NumPyro; causal DAGs and treatment-effect estimation; agent-based simulation and synthetic populations; probabilistic forecasting. Models that say how sure they are, and about what.
ai
Applied LLM systems with evaluation and scoring pipelines built in, adaptive question engines, and machine learning that ships. The model is rarely the hard part; the system around it that keeps it honest is.
selected work
built and shipped
Three systems from the practice, each one a working version of the demos above at production scale.
consumer health · causal inference
serif
A consumer-health causal inference engine that turns wearable and biomarker data into personalized, dose-aware recommendations through a causal DAG and Bayesian posteriors.
1,188 participants · 3 cohorts
adaptive measurement · bayesian
prism
A Bayesian adaptive political-disposition quiz built on 14 latent dimensions and 130 archetypes. Questions adapt in real time to what each answer reveals.
converges in about 30 questions
prediction markets · forecasting
markets
Prediction-market tooling: Bayesian decomposition of contract prices into shared latent terms, live dashboards, and an active forecasting track record.
prices → shared latent terms
the background
credentials
The demos are the argument; the paper trail is here for completeness.
- B.A. Economics & MathematicsNYU · 2011
- M.B.A.Columbia Business School · 2019
- M.S. Computer ScienceGeorgia Tech · 2024
Fifteen years across quantitative finance, data science, and software engineering, most of it spent turning uncertain data into decisions someone had to stand behind.
get in touch
bring a decision worth making well
The fastest way to find out whether this practice fits your problem is thirty minutes on a call. Bring the decision you are trying to make; leave with a sharper version of it.
elsewhere: linkedin.com/in/samcialek · essays at matricesofconfusion.com