Principal Engineer at FarEye · 9+ years experience
I turn messy logistics problems into systems that ship.
9+ years across route optimisation, geospatial services, AI agents and full-stack platforms. As Principal Engineer at FarEye, I own the routing engine end to end. Here I write about solvers, systems and AI-native engineering.
Results
- 6 h → 20 min
- Route planning time with Pilot Core, the AI dispatch agent
- −53%
- Monthly cloud spend after moving from EC2 to Kubernetes ($17K → $8K)
- $70K / yr
- Map API spend removed by self-hosting OSRM and Valhalla
- 7 regions
- Where the VRP solver runs in production
Writing
All posts →- Primer: probability and statistics
Random numbers, seeds, expectation, spread and percentiles: the statistics you need to read randomized routing results and compare methods fairly.
- Neighborhoods and the 2-opt move
How local search improves a tour one small change at a time: the 2-opt move, neighborhoods, delta evaluation, and how far 2-opt gets on berlin52.
- Approximation algorithms
How the double-tree and Christofides algorithms guarantee a tour within 2× and 1.5× of optimal, why that needs the triangle inequality, and its limits.
- Construction heuristics
Build a tour in milliseconds with nearest neighbour, three insertion rules and a space-filling curve, and measure each against the berlin52 optimum.
Selected work
All projects →- FarEye's Vehicle Routing Solver
The route optimisation engine behind FarEye's planning product, built on Iterated Local Search and running in seven regions.
- Pilot Core, an AI dispatch agent
An AI agent that runs a dispatcher's planning workflow from start to finish.
- Self-hosted routing engines
A build-vs-buy analysis that replaced paid distance-matrix and directions APIs with self-hosted OSRM and Valhalla.
- FarEye Track
A real-time shipment visibility platform built on Kafka and Elasticsearch, with ML-predicted delivery dates and threshold alerts.