← Work
2025-07

Cross-Border Fleet Optimizer

A super-opinionated GPS for a tiny cross-border fleet, optimized for cost not just distance.

logisticspythonoptimization

Reality check: not a full TMS. But it already plans better than a stressed human with Google Maps, a whiteboard, and vibes.

Quick scan
What it is A Python tool that plans multi-truck routes with demand and time windows, outputs a route plan and a map, and minimizes total cost (distance + driver time) rather than pure kilometers.
Why it matters Border logistics is huge and weirdly low-tech. Humans are forced to do combinatorial planning manually, under stress, with poor cost visibility.
Payoff Once cost + real travel times were wired in, the tool started making non-obvious but believable decisions (longer distance, cheaper total cost). That’s when it became a real tool.

What this actually is

You give it a depot, a set of cities, demand per city, and simple delivery windows.

It outputs: which stops each truck visits, the order, rough arrival timing, and a map that makes the day obvious.

It’s meant for the person planning routes with a whiteboard, Google Maps, and intuition.

Why I bothered to build it

Humans are bad at this class of problems: too many combinations, messy constraints, and tradeoffs between time, distance, and cost.

I wanted to see if I could build something that plans better than a human while still being simple enough to trust.

What actually worked

  • Cost-based routing produced good non-obvious routes (not just shortest distance).
  • Switching from straight-line distances to real road data changed everything.
  • One-command scenario runs made testing fast (CSV in → plan out).
  • Map output tells the story faster than any table.

What is still janky

  • Texas routes are real (OSRM). Mexico side still has approximation gaps depending on coverage.
  • Cost model is simple (flat per-km + per-hour; no tolls, overtime, fuel differences).
  • Time windows are hard constraints; no ‘least bad’ soft penalties yet.
  • Setup is not dispatcher-friendly (Docker + OSRM + patience).

What I would do next

  • Full-region routing data for true cross-border realism.
  • Richer cost model (overtime, tolls, country-specific fuel).
  • Soft time-window penalties (late at a cost instead of solver failure).
  • Simple web UI (upload CSV → run → map + report download).

Technical appendix

  • Problem type: CVRPTW (Capacitated Vehicle Routing Problem with Time Windows)
  • Language: Python
  • Libraries: pandas (data), OR-Tools (routing), folium (maps), geopy (fallback distances)
  • Routing: local OSRM in Docker for realistic travel times/distances; straight-line fallback when needed
  • Objective: minimize total cost = distance × cost_per_km + time × cost_per_hour
  • Design detail: routing logic separated so OSRM can be swapped without touching optimization core; timestamped outputs for traceability

← Back to Work