Our algorithms

Operations research methods tuned for real‑world logistics and industrial engineering.

Linear & mixed‑integer programming

We frame core planning as LP/MIP: assignment, knapsack, cutting‑stock, and facility‑location variants. Solved with efficient branch‑and‑bound and cutting planes on bounded problem sizes.

Vehicle routing heuristics

For VRP with time windows and capacities, we use constructive methods (sweep, savings), then improve with local search (2‑opt, 3‑opt), large neighborhood search, and guided tabu search.

Bin‑packing & load building

We combine first‑fit/ best‑fit decreasing with geometric checks to respect weight distribution, cube limits, and stacking rules for pallets, trailers, containers, and ULDs.

Scheduling & time windows

Greedy and dynamic programming to schedule docks, appointments, and crew within service windows, with penalties for early/late arrivals and soft constraints when needed.

Multi‑objective scoring

Objective blends cost, distance, lateness, utilization, and CO₂. We normalize, weight per user preferences, and use Pareto‑efficient moves during search to balance trade‑offs.

Robustness & uncertainty

Buffers for dwell and traffic variability; scenario sampling to keep plans stable under realistic fluctuations without being overly conservative.

Explainability

Every plan exposes the constraint set, scoring breakdown, and the improving moves applied, so planners can audit and adjust with confidence.

Performance on the edge

Compute budgets are tuned for fast responses on Cloudflare Workers. We cap instance sizes, use incremental search, and cache reusable subresults where appropriate.

Working principle

Dive into our LP/MIP models, VRP heuristics, and edge architecture.

View working principle