Dynamic Route Optimizer
Reinforcement learning route planner that adapts to live traffic, weather, and SLA tiers.
- Client
- Helios Logistics
- Year
- 2024
- Duration
- 14 weeks
- Team
- 4 engineers, 1 operations research specialist
Helios dispatchers were building routes at 6am that were materially wrong by 9am. A previous optimizer had been switched off because dispatchers could not tell why it was proposing what it proposed. We rebuilt it around continuous replanning with explainable recommendations, and dispatcher trust is what made it stick.
The brief
Replan routes continuously against live conditions, honoring hard constraints, in a way dispatchers will leave running rather than override.
- Logistics
- RL
- Optimization
The decisions that mattered.
Every engagement turns on a handful of calls. These are the ones that decided whether this system reached production.
Explainability was the adoption blocker
Every recommendation shows the constraints it honored and what it traded off. Dispatchers audit a suggestion in seconds instead of guessing, which is why override rates fell rather than climbed.
Hard constraints stay hard
Hours-of-service, vehicle capability, hazmat restrictions, and SLA tiers are modeled as constraints the optimizer cannot violate — not as penalty terms it can trade away.
Calibrated ETAs, not optimistic ones
Arrival windows carry validated confidence intervals. An 80% window contains the actual arrival 80% of the time, which is what made customer-facing ETAs safe to publish.
Offline-first driver application
Coverage gaps are routine on long-haul lanes. The driver app works fully offline and reconciles on reconnect, so a dead zone never loses a proof of delivery.
What shipped, and what changed.
Outcomes
- 1.4M routes optimized in the first year
- 8,200 vehicles orchestrated daily at steady state
- ETA accuracy improved 22 percentage points
- Dispatcher override rate down from 34% to 7%
“The difference from the last system is that our dispatchers can see the reasoning. That is the entire reason this one is still switched on.”
What we built it with.
- Python
- PyTorch
- OR-Tools
- Kafka
- Postgres
- React Native
Other projects from the shelf.
Clinical Documentation Copilot
Ambient AI scribe that drafts structured notes, ICD-10 codes, and prior-auth letters in real time.
Read moreEquity Research Terminal
Multi-modal RAG over 10 years of filings, transcripts, and broker research with citation-grounded answers.
Read morePredictive Maintenance Platform
Time-series + vision models predicting equipment failure 72 hours in advance across 14 plants.
Read moreTell us about the problem.
A 30-minute call with the senior team. We will tell you what is realistic, what it would cost, and whether we are the right people for it.