matrices is a consulting practice that builds infrastructure for adaptive decision-making.
our clients share a shape: a decision with a date on it, data nobody has had time to evaluate, and a skeptical room (an allocator, an auditor, a diligence analyst, an investment committee) waiting to attack the answer. we build answers that survive that room: systems that update beliefs as new data arrives, quantify causal effects from observational data, and simulate complex behavior at population scale.
founded by sam cialek, the practice draws on 15 years of experience in quantitative finance, data science, and software engineering. past and current engagements include consumer health AI, sustainability analytics, political modeling, and hedge fund research infrastructure.
degrees
certifications
research leads with a shelved question: a dataset nobody has had time to evaluate, backtests that don’t match live, headcount frozen until next year. we build the first version, hand over documented code, and say straight whether the thing is worth anything; the bayesian workflow is why the verdict survives your own quants.
owners and asset managers with NYC or california exposure and no sustainability data staff, facing LL97, SB253, or a GRESB deadline. we deliver the audit-ready scope 1/2/3 baseline, the filing itself, and the retrofit order; scenario modeling turns compliance into a capital plan.
founders and science leads sitting on wearable, lab, and engagement data whose recommendations are still if-then rules. we design causal recommendation engines and investor-grade outcomes analyses, honestly caveated; we shipped an engine covering 1,188 participants across 3 cohorts.
product leads whose fund clients want restructuring filings, holder disclosures, and dockets as structured, queryable data. we design the canonical schema and the LLM-assisted extraction pipeline with human review; confidence scores decide which documents route to a person.
litigation finance, specialty insurance, corporate development, event-markets desks: teams that must put one defensible number on a novel event. we decompose it into a sourced decision tree and propagate the uncertainty by simulation, so the committee sees which assumption drives the error bar.