How we approach real problems.

01 · Problem

A regional logistics operator tracks shipments across several carrier portals, a warehouse system and shared spreadsheets. Delays are discovered when customers call, and staff spend hours each day chasing status updates.

02 · What We Built

An event-driven data platform that ingests carrier APIs, EDI messages and IoT telemetry into one live view. A machine-learning model predicts arrival times, and an AI agent drafts customer updates for exceptions, which staff approve before sending.

03 · Technology

04 · Result

Designed to give operations one real-time view of every shipment, surface delays hours before customers notice, and remove most manual status chasing.

Concept project: result describes the design target, not a measured client outcome.

01 · Problem

Staff at a professional services firm search shared drives, the intranet and long PDF manuals to answer routine questions. Answers are inconsistent, and compliance is wary of any AI tool that can't show its sources.

02 · What We Built

A retrieval-augmented generation (RAG) assistant that respects existing document permissions, answers in English or Traditional Chinese, and cites the source paragraph. An evaluation set of real staff questions measures accuracy before each release. The assistant runs inside Microsoft Teams.

03 · Technology

04 · Result

The prototype returns cited answers in seconds. It is built so accuracy can be measured and signed off by compliance before rollout, and so staff only see content they already have access to.

Concept project: result describes the design target, not a measured client outcome.

01 · Problem

An accounts payable team keys thousands of supplier invoices into the ERP each month from PDFs, scans and photos. Manual entry causes errors, duplicate payments and late-payment penalties.

02 · What We Built

A document pipeline combining OCR with LLM extraction, then validation against purchase orders and the vendor master. Each field gets a confidence score, and low-confidence invoices go to a review queue. An agent emails suppliers about missing information, and approved invoices are posted through the ERP API.

03 · Technology

04 · Result

Designed so clean invoices flow straight through without manual keying, people review only the exceptions, and every decision has an audit trail.

Concept project: result describes the design target, not a measured client outcome.

01 · Problem

Sales reps at a B2B distributor manage customers through WhatsApp and personal spreadsheets. Management has no pipeline visibility, and quotes take days because stock and pricing sit in the ERP.

02 · What We Built

A custom CRM with accounts, pipeline and quoting linked to live ERP stock and price lists. A WhatsApp Business integration logs conversations against the customer record. AI summarises calls and chats and suggests next steps, and a mobile app lets reps work from anywhere.

03 · Technology

04 · Result

Designed to give management a single view of every customer and the pipeline, and to cut quote turnaround from days to the same day.

Concept project: result describes the design target, not a measured client outcome.

01 · Problem

A freight software vendor installs a separate copy of its system for every client. Onboarding takes weeks, upgrades are painful and hosting costs grow with each new customer.

02 · What We Built

A multi-tenant SaaS platform with tenant isolation at the database level, per-tenant configuration and branding, booking and document management, shipment tracking, a customer portal, SSO and subscription billing.

03 · Technology

04 · Result

Designed so a new operator can be onboarded in days instead of weeks, every customer runs the latest version, and hosting cost per tenant is predictable.

Concept project: result describes the design target, not a measured client outcome.

01 · Problem

A retail brand runs separate systems for its website, mobile app and stores. Inventory is out of sync, promotions differ by channel and the website is slow on mobile.

02 · What We Built

A headless commerce architecture with shared product, inventory and pricing APIs feeding a Next.js storefront, a React Native app and store POS. It includes personalised recommendations and local payment methods.

03 · Technology

04 · Result

Designed to keep stock and promotions consistent on every channel, with a storefront built to pass Core Web Vitals on mobile.

Concept project: result describes the design target, not a measured client outcome.