Industries Logistics & Supply Chain
Where one delay movesthe whole operation.
AI Engineering and software development for logistics companies.
Carriers, warehouses, brokers, and customers all need the same picture at the same time, and routing has to hold when one leg slips. That is where AI does real work, and it is the layer we build.
OptiwayRoute Optimization for Last Mile Delivery
Route optimization software for last mile delivery: a dispatch platform that sequences a day of stops in one click, and a driver app that carries them to the door. We worked on it as an embedded engineer inside the client's own team, on the live route predictability algorithm.
- 100K+
- Customers on the platform
- 5M+
- Deliveries optimized
- 25%
- Lower delivery costs
Why logistics companiestrust us with delivery.
A carrier that stops sending updates, or a customs form with one wrong code, ends in detention charges and a team on the phone all afternoon. This is why we pay special attention to the most sensitive parts of the system.
Integration is the industry
If your last visibility project ended in a dashboard nobody trusts, the feeds are why. EDI messages, carrier portals with no API, and PDFs emailed by partners have to become one shipment record first.
Exceptions are the actual job
Most shipments need nobody's attention, and a system that shows all of them buries the few that do. The build has to surface the ones going wrong while there is still time to act.
Optimal on paper, unrunnable in practice
A route plan your dispatchers override on day one was never optimal. Driver hours, dock windows, temperature rules, and the customer who refuses a 6am delivery have to be constraints in the model.
Documents move the freight
One mistyped code on a customs declaration holds a container at the port at a daily rate. Clearance speed is set by the worst document in the file, which is what extraction has to be built for.
Forecasting after the stable years ended
Safety stock set on a calm decade is either capital sitting still or a stockout waiting for the next disruption. A forecast has to be fitted on the demand and lead-time variability your lanes show now.
What logistics demandsfrom engineering and AI.
01Visibility that ends in a decision
A control tower showing every shipment is a wall of green nobody reads. The useful version ranks the exceptions by consequence and attaches the recommended action and the person who owns it, so seeing a problem and acting on it become the same moment.
02ETAs people can plan against
Downstream labour and dock schedules get planned on arrival times, and carrier-provided ETAs are optimistic and rarely updated. Predictions worth planning against are built on your own lane history, current dwell, weather, and port congestion.
03Document automation through to clearance
Customs clearance moves at the speed of the paperwork, and every hour lost there is priced in demurrage. Meeting it takes extraction, classification, HS code assignment, and validation against destination rules, with confidence scoring that routes uncertain cases to a broker instead of guessing.
04Network design and inventory positioning
Where stock sits determines what service level is even possible to promise. Answering that means modelling the whole network, its nodes, lanes, cost to serve, and the trade-off between working capital and speed, where a spreadsheet can only approximate one scenario at a time.
Words from the front lines of Logistics.
Get In Touch
Inside every logistics platform you're juggling between systems you do not own: a carrier feed, a data provider, a payments service, the notification pipelines. Every one of them has failure modes you do not control, so you design everything to degrade gracefully and heal itself instead of assuming the happy path. The self-healing jobs and the strict test discipline are what make a launch calm instead of chaotic.
Nortik’s Impact
Integration-heavy platform delivery on a fixed timeline: provider pipelines, payments, and notification flows built to degrade gracefully and recover on their own, behind the test discipline that keeps a launch predictable.
Discover more of our insights.
Recent Posts
What an AI agent actually costs to run in production
Token bills are the visible line item. The real costs hide in the architecture: retries, tool calls, human review loops. A breakdown from live systems.
Why AI prototypes fail in production (and how to ship one that doesn't)
Most AI prototypes stall somewhere between the demo and the deploy. The problem is rarely the model. It's the evals, guardrails, and infrastructure nobody scoped.
The eval suite is the product: testing LLM features like an engineer
Prompt changes without a test harness are blind bets. How we build eval suites from real traffic, and why they outlive every model you'll ever swap in.
Frequently asked questions
Everything operators usually want to know before building on their shipment data. Not here? Ask us directly.
Yes, along with the EDI layer underneath them: 214s, 856s, 990s and the rest, plus the carriers who still communicate by portal, email, or file drop.
We build the normalisation layer so those sources produce one shipment record, rather than asking your partners to change how they send you data.
No, but it does set the ceiling, so we measure it openly: coverage and freshness per carrier, per lane, made visible rather than averaged away.
Where a carrier's data is genuinely unusable, the model says so instead of producing a confident ETA out of nothing. A known gap is manageable; an invented number isn't.
In most networks, meaningfully. The baseline is a carrier estimate that ignores your own history on that lane, current dwell, and seasonal congestion, so there is usually real room to close.
We back-test against your delivered history before anything goes live, so the improvement is a measured number rather than a promise.
That is the work we did on Optiway, embedded in the client's own team on the live route predictability algorithm behind its sequencing and arrival times.
We automate the document work: extraction, classification, HS code suggestion, and validation, with confidence thresholds that route anything uncertain to a human broker.
Denied-party screening and the filing itself stay with the licensed parties who are accountable for them. We make their work faster and better evidenced, not automatic.
Rates, volumes, and customer lists are the most sensitive assets in this business, so access boundaries and tenancy separation are designed before anything is built.
Nothing is pooled across clients to train a shared model unless you explicitly ask for it and the terms make that possible.
You do. Code, trained models, pipelines, and documentation are yours, along with the infrastructure access to run them without us.
We hand over a codebase your own team can pick up. Nothing we build is designed to keep you dependent on us.