Algormind is a decision support practice. We build mixed-integer optimization models that set which routes are run by which contracted provider or internal division, from which depots, and with which vehicle technology, calibrated to each agency's own data and operating rules.
The work sits one layer above scheduling, and the allocation it produces becomes the starting point for the scheduling tools an agency already runs.
Start a conversationWe take the allocation decision that transit agencies make continuously and treat it as a formal optimization problem: minimize avoidable operating cost, mainly deadhead, while holding service levels, contractual shares, fleet availability and depot capacity as constraints.
Mixed-integer models that assign service to providers, depots and vehicle technologies under the operating rules that actually bind.
Baseline against optimized alternatives on identical service, priced in the agency's own cost categories and reported in the terms it already files federally.
Vehicle range, depot technology compatibility and charging capacity enter as constraints on the same assignment decisions.
Commercial scheduling platforms optimize crew and vehicle assignments within an allocation that has already been set, and they take that allocation as given. In many systems the allocation itself is revisited through manual judgment or through one-time studies as conditions change, without a recurring formal optimization step.
That is the layer we work at. We complement scheduling software rather than replace it, and the models are built for a specific system, not sold as packaged software.
The cost of an allocation that has drifted shows up mainly as deadhead, the closed-door movement that carries no passengers. Deadhead is reported to the National Transit Database, so the size of an agency's exposure can be established from public filings before any engagement begins. What portion of that exposure is realizable depends on each agency's variable cost structure, labor arrangements and maintenance profile.
We build an opportunity analysis from the agency's own published NTD filings: deadhead share against the national average, what the gap represents at the agency's own reported cost per vehicle mile, and where a formal allocation step would be worth running.
Under an engagement, the model is calibrated to internal operating data and reconciled against the current allocation before any scenario is produced, so differences surface and are resolved with planning staff.
Alternatives covering identical service, each stating the cost difference and the operational change it requires, reviewed with agency staff before any decision.
Conditions change: contracts, fleet mix, depots, demand. A calibrated model is re-run on a regular cadence rather than left as a one-time study.
Ten years of implementations of this class of model in bus and bus rapid transit systems in Bogotá and Cali, Colombia, for contracted providers and for a public transport authority, from January 2013 to February 2023. The models are built in GAMS and solved with commercial solvers.
Client names, engagement documents and detailed results are available on request, subject to the confidentiality terms of each engagement.
If you operate or plan bus service and want to know what your allocation is costing you, the first analysis is built from public data and costs you nothing.
Alain Baquero Ortegón, Founder
Operations Research Analyst applied to Public Transportation
New York, NY