ORDERS
1,284
IN TRANSIT
342
ON-TIME %
94.2
Background
Logistics dispatch is a combinatorial problem of enormous practical complexity. A dispatcher managing a fleet of 200 vehicles, thousands of daily orders, and hundreds of real-world constraints—time windows, cold chain requirements, driver skill qualifications, restricted zones, split deliveries—cannot produce optimal routes manually. They produce feasible routes, usually ones that vaguely resemble yesterday's routes. The difference between feasible and optimal translates directly to wasted fuel, underutilized capacity, and missed SLAs.
Our client's operations team was running dispatch out of Excel spreadsheets, making decisions reactively as new orders arrived or vehicles broke down. We built Dispatch Commander—a Vehicle Routing Problem (VRP) solver that models millions of constraints, computes provably optimal routing decisions, and pushes dispatch actions directly to the operations system in real time. It re-optimizes continuously as conditions change.
The Challenges
- •Manual dispatch decisions could not keep pace with real-time order changes, traffic disruptions, and vehicle breakdowns—every reactive decision degraded the global plan
- •Suboptimal routing was costing the client an estimated 6–10% in avoidable fleet operating cost per month
- •The constraint space was genuinely hard: time windows, priority SLAs, multi-depot pickup/delivery, cold chain requirements, crew skill matching, and restricted delivery zones all applied simultaneously
- •No audit trail existed for dispatch decisions—finance and compliance had no way to verify that routing choices were defensible
- •Replanning after a disruption took dispatchers 30–45 minutes of manual rework per incident
Our Approach
We built Dispatch Commander in three layers: a comprehensive constraint model that captures real operational reality, a deterministic optimization engine that produces provably optimal plans, and an audit-ready evidence chain that makes every decision traceable.
Phase 01: Constraint Modeling at Production Scale
Before writing a solver, we spent significant time modeling the client's real operational constraints accurately. The constraint model covers: time windows and priority SLAs, multi-depot pickup and delivery sequences, vehicle capacity and load type compatibility, cold chain temperature requirements, crew skill-to-task matching, restricted delivery zones and road network constraints, and split/merge delivery rules. Modeling these constraints correctly is what separates a solver that works in demos from one that works in production.
- •40+ constraint types modeled: time windows, capacity, cold chain, crew skills, zone restrictions
- •Real-time disruption inputs: traffic updates, vehicle breakdowns, order additions/cancellations
- •Safe dispatch with rationale attached—every action includes the constraints that drove it
Phase 02: Deterministic VRP Optimization Engine
The solver uses a hybrid approach: mixed-integer programming (MIP) for exact solutions on tractable subproblems, combined with constraint-based local search for real-time re-optimization at scale. The execution loop is continuous: new orders and telemetry feed in, the solver re-optimizes the global plan, and updated dispatch actions are pushed to the operations system—typically within seconds of a change event. The solver doesn't just find a good solution—it finds the provably optimal one under the current constraint set.
- •Hybrid MIP + constraint-based local search for optimal routing under millions of constraints
- •Sub-second re-optimization on incremental changes (new order, breakdown, traffic update)
- •Direct API integration with operations and dispatch systems—no manual transcription
Phase 03: Audit-Ready Evidence Chain
Every routing plan generated by Dispatch Commander includes a full evidence record: the constraint set active at the time of the decision, the optimality rationale, a versioned snapshot of the route plan, and KPI projections. We built an evidence export that translates technical routing decisions into plain-English summaries for finance and compliance review—enabling the operations team to defend any routing decision to auditors without needing an engineer in the room.
- •Immutable plan snapshots: every route plan versioned with timestamp, constraints, and KPI projections
- •Evidence pack generator: human-readable summaries of routing decisions for finance and audit
- •Versioned rollback: any previous plan can be restored as the active dispatch in under 60 seconds
The Results
50% reduction in dispatch labor—the same team handles significantly higher order volume
6% reduction in fleet operating cost from more efficient routing and capacity utilization
5% reduction in total miles driven across the fleet
Replanning time after a disruption dropped from 30–45 minutes to under 30 seconds
Full audit trail for every routing decision—finance and compliance reviews now take hours, not days
Final Takeaway
The gap between feasible and optimal in logistics is worth real money—and it compounds daily. The reason most companies stay stuck with manual dispatch isn't that they don't know it's suboptimal; it's that building a VRP solver that handles real-world constraints at production scale is genuinely hard. Constraint modeling, solver architecture, real-time integration, and audit-ready evidence all have to work together. We built a system that does all four.
Technologies We Use
Modern, proven technologies to build robust applications
Python
OR-Tools (VRP Solver)
MIP (Mixed-Integer Programming)
Kubernetes
Kafka
PostgreSQL
React
See More Work
passport · I-129 · DS-160
policy KB + 12,400 cases
cross-checking eligibility
pending
easyvisas.ai – Fully AI-Powered Immigration Processing
Agentic AI backend with RAG over immigration policy and 12,000+ historical cases — automated document intake through decision support.
Mipmap3D Platform
Cloud-native 3D reconstruction platform for surveyors and construction, delivering 60–80% cost reduction.