AI agents built for the pace of a transshipment hub

workflows: mapped

What actually slows a Singapore freight operation

A transshipment operation produces documents faster than people can read them: bills of lading, arrival notices, delivery orders, commercial invoices, packing lists, declarations prepared for TradeNet. Order intake lands overnight from US and EU customers, so the queue is longest while your ops team is asleep. The data lives in CargoWise, SAP, a TMS or WMS — and, more than anyone admits, in Outlook threads and WhatsApp chats with hauliers. The bottleneck is rarely transport capacity. It is the keying, matching, and chasing between systems, done by people who should be handling exceptions instead.

agents: in-production

The four agent patterns that fit this industry

Four patterns cover most of a freight operation's manual load. Each deploys narrow first — one document set, one trade lane — then widens once the eval numbers hold. Claude-first for document-heavy reasoning, model-agnostic wherever the benchmark says otherwise.

• Order intake — parses customer POs and booking requests from email into structured orders, whatever time zone they arrive from
• Document extraction — bills of lading, arrival notices, and packing lists validated against the live shipment record
• Three-way matching — PO, proof of delivery, and freight invoice reconciled line-by-line, with exceptions routed
• Exception triage — rolled containers, customs holds, and missed slots classified, drafted, and escalated by severity

pdpa: by-design

Compliance designed in, not bolted on

Shipment documents carry personal data — consignee names, addresses, contact numbers — so Singapore's PDPA, administered by the PDPC, applies to every agent that touches them. We design deployments against PDPC expectations: data minimisation and redaction before model calls where required, defined retention rules, region-appropriate hosting options, and a full audit trail of every extraction and action. If your Singapore front office shares data with a Malaysia back office, cross-border transfer obligations are part of the architecture from day one. Human approval gates sit on anything that posts money or commits capacity.

evals: gated

Nothing ships without passing its eval

Every build runs through our Agent GPA evaluation before go-live — scored against your real documents, not demo PDFs. An agent that misreads a bill of lading fails the gate and gets fixed before it ever touches production volume. Behind the gate sits the operations record: 300+ production incidents handled, 30+ resilience assessments, and 50+ load-test audits. After go-live, agents run in shadow mode on live volume first, earn autonomy workflow by workflow, and stay monitored the way we monitor any production system.

rollout: sg+my

Getting started, one workflow at a time

We start with a scoping call, not a platform pitch. Pick the workflow that hurts most — freight document intake, three-way matching, or exception triage — and we map the systems it touches, define the eval gate, and quote fixed in SGD. First production agent in weeks, not quarters. Regional operations get one more advantage: the same Suntec-based team runs deployments across Singapore and Malaysia on one architecture, so an SG+MY rollout means one pattern and one accountable team, not two vendors reconciling their differences.