Interactive demos

Real solvers, running live in your browser

Interactive demos on real problems — a GTFS transit network optimized end to end, exact layout under a separation rule, a continuous robustness measure, and a metaheuristic failure taken apart against a proven optimum. Each has a guided walkthrough of the theory, how the instance is built, how to read the result, and what we'd do for you.

Assignment · rules in plain English

Optimization over language

Assign real papers to reviewers where the objective — and the reviewers' rules, written in plain English — are read from the text by a calibrated model, then optimized exactly in your browser. Toggle a rule and watch it re-solve.

Open demo
Team orienteering · LNS · live

311 dispatch — urgency read from the text

A real backlog of open Atlanta 311 requests, triaged by an AI that reads each citizen's own words for danger, then routed live in your browser. Toggle between reading the text and the usual oldest-first rule and watch which hazards get cleared.

Open demo
Real Python · sandboxed · show-code

Grounded — AI data analyst

Ask a spreadsheet a question in plain English and get a chart, a table, and a written interpretation — computed by real Python run in an isolated sandbox, never guessed. A frozen replay of the real engine, with the generated code on a toggle.

Open demo
CVRP · Split · SWAP* · live HGS

Vehicle routing (Hybrid Genetic Search)

A capacitated VRP solved with a compact Hybrid Genetic Search — a giant tour, the Split decode, SWAP* local search, and biased-fitness evolution, all in your browser. Built from Thibaut Vidal's research.

Open demo
GTFS · min fleet + crews

Transit network optimizer

Turn a real MARTA bus feed into a plan — minimum fleet (exact), driver duties, and a weekly roster. Or upload your own GTFS.

Open demo
Max independent set · exact

Separation-constrained layout

Fit the most people into a room under a minimum-separation rule — solved to proven optimality, live in your browser.

Open demo
Interval uncertainty · η ∈ [0,1]

The price of robustness

At identical spend, which plan best resists uncertainty? A robustness score computed two ways that agree.

Open demo
Simulated annealing · proven optimum

Anatomy of a failed anneal

201 committed runs, none feasible. Take the failure apart, then reach a few percent of the exact optimum in milliseconds.

Open demo