LUMUT, PERAK — 10 September 2026 — Quiver Dive Team Lumut Sdn. Bhd. today announced an initiative to build an industry AI cloud platform serving logistics, freight forwarding, customs brokerage, cross-border trading and warehousing customers in Malaysia and the wider Southeast Asian region.
The platform is organised in four layers: a data and knowledge layer covering industry corpora, document samples, handling imagery and operational records; a model layer combining general foundation models with domain-specific models; a capability layer exposing agent orchestration, visual recognition, document parsing and forecasting as reusable building blocks; and a service layer delivering open APIs, a customer console, token metering and private or hybrid deployment.
Three services will be offered to industry customers. A fine-tuning and training service covers data assessment and cleaning, annotation standards, supervised fine-tuning and parameter-efficient adaptation, evaluation against a held-out set agreed with the customer, knowledge injection through retrieval, and ongoing iteration with version-level rollback. An AI-agent service cloud provides hosted, task-oriented agents for quotation and enquiry, document processing, shipment tracking, customer service desk, safety inspection and supply chain analysis, with permissions, audit trail, canary release and a human-in-the-loop takeover path. An AI-token cloud provides a unified model gateway with per-project quotas, cost dashboards and budget alerts, caching and routing optimisation, call-level logging and rate limiting.
The initial model hub covers logistics document recognition, HS code and declaration assistance, freight rate and capacity forecasting, cargo visual condition recognition, a multilingual customer service model for English, Bahasa Melayu and Chinese, and warehouse safety behaviour recognition. Each model is published with its applicable scenario, input and output, delivery form and current maturity, stated as live, in pilot or planned.
The company confirmed that customer data is isolated per tenant with least-privilege access, that imagery, documents and call logs are retained only for an agreed period, and that personal data handling follows Malaysian personal data protection requirements. Model output is positioned as decision support: declarations, quotations and safety findings remain subject to confirmation by a responsible officer, and no model is presented as delivering absolute accuracy.
Engagement follows a staged path — capability trial in a sandbox, a bounded scenario pilot with success criteria agreed in advance, joint fine-tuning, then a platform subscription. Interested parties can request a capability session through the contact page on this website.