$ matrices --init
matrices :: independent practice :: automation · optimization · modeling · ai
matrices is the independent practice of sam cialek: fifteen years across quantitative finance, data science, and software engineering, focused on building systems that run unattended and get better while they run.
# four capabilities, one practice
Agentic pipelines, workflow orchestration, and LLM-powered systems designed to run unattended. Human-in-the-loop checkpoints sit exactly where judgment matters, so the system does the routine work and people make the calls that count.
Portfolio and resource-allocation optimization, and decision analysis under uncertainty. Objectives are distribution-aware, not just expected value: the tails, the variance, and the cost of being wrong all belong in the objective.
Bayesian inference in Stan, PyMC, and NumPyro. Causal DAGs and treatment-effect estimation, agent-based simulation with synthetic populations, and probabilistic forecasting. Models that state their uncertainty instead of hiding it.
Applied LLM systems that ship: evaluation and scoring pipelines, adaptive question engines, and machine learning embedded in real products. Built to be measured, so they can be trusted.
# selected work
A consumer-health causal inference engine. Serif turns wearable and biomarker data into personalized, dose-aware recommendations by propagating each user's data through a causal DAG and reading actions off Bayesian posteriors.
└─ 1,188 participants · 3 cohorts · causal DAG + Bayesian posteriors
A Bayesian adaptive political-disposition quiz. Prism models 14 latent dimensions and matches respondents against 130 archetypes, choosing each question in real time based on what it has already learned.
└─ 14 latent dimensions · 130 archetypes · converges in about 30 questions
Prediction-market tooling. Bayesian decomposition of contract prices into shared latent terms, so related contracts inform each other, with live dashboards and an active forecasting track record behind it.
└─ shared latent terms · live dashboards · active track record
# operator
# open a channel
If you have a decision system worth building, or a process that should be running itself by now, get in touch. A 30-minute conversation is usually enough to tell whether there is a fit.