Every demo compares a sensible baseline with the model and shows the result in one line. Optimization and causal models are solved live on a Python server; analytics, forecasting, experimentation and GenAI demos compute in your browser, and the GenAI ones replay pre-computed LLM steps. No company data, identifiers or prompts are used.
First-come cart vs the cart with the fewest aisles and an exact route; whole-wave trip packing and dispatch.
runs in your browserTwo ring conveyors replay the same batch: arrival order vs the order found by simulation and Tabu Search.
Python server · 2–8 sStock-where-it-sells vs optimized storage, sourcing and pooled safety stock under capacity.
Python server · MILPA rule-of-thumb day vs a CP-SAT schedule that respects every operating rule and the target seat share.
Python server · CP-SAT ≈ 8 sCorrelation screen vs order-constrained causal discovery, then do-interventions on yield.
Python server · ≈ 1 sThree businesses with no shared codes, one interest language, and cohorts read across them.
runs in your browserKeyword filter vs retrieval + LLM judgment vs independent verification, on two invented ads.
browser · LLM steps replayedA first draft with an invented claim, caught by the verification layer and regenerated.
browser · LLM steps replayedGeneric chatbot vs a report-grounded, cited answer; a guardrail on treatment questions.
browser · LLM steps replayedSpecies-only vs species plus protein-family clusters, tested on held-out cohorts; a cluster index searching a protein catalogue.
runs in your browser · ≈1–2 sLast week's sales plus 10% vs a global forecast across 300 stores with production at the profit-maximizing quantile.
runs in your browserMonth-to-date run-rate vs a daily model with a probability of hitting the target and a traffic light a month ahead.
runs in your browserCatalogue-average curve vs feature total + Bass curve for a new title; seasonal naive vs an additive model for the market.
runs in your browserCorrelation ranking vs a PCMCI lagged causal graph, with each lead's effect against the planted truth.
runs in your browserColumn shuffling vs a Gaussian-copula generator under the same report: marginals, correlations, utility, privacy.
runs in your browserA confounded before/after launch vs a planned randomized test, Bayesian probability and Thompson sampling.
runs in your browserBilling-account grouping vs Louvain communities and strong-tie households on calls, night location and accounts; IDs tracked a month later.
runs in your browserToday's bill carried forward vs survival × revenue trajectory, checked against realized 24-month value.
runs in your browserRaw usage rank vs RFM levels binned by a genetic algorithm and weighted by least squares; churn and ARPU by decile.
runs in your browser