Blog

Most organisations measure whether trainees liked a course; few measure whether results changed. A digital system for the Kirkpatrick model is mostly a measurement plan per programme, timed surveys and supervisor checks, reminders, one results database and a link to the operational KPIs. AI helps as a supporting text-analysis layer, with people reviewing every output. Sized honestly, that is three servers and one GPU, not an AI platform.

A GB300 desktop and a small DGX Spark cluster do different jobs. The GB300 has 252 GB of memory at 7.1 TB/s and serves large models to a whole team, split into up to seven slices. A Spark has 128 GB at 273 GB/s: enough to fine-tune a 70-billion-parameter model, too slow to serve one to many people. Put serving on the GB300 and development, fine-tuning and tests on the Sparks. The graph, rules and search software around the models runs on CPUs and needs its own tier. Prices and export rules for the UAE are at the end, checked on 16 October 2026.

Every data vendor a commodity trading desk already pays for now ships its own AI chat. What nobody sells is one assistant across the CTRM, the mailboxes, the contracts, the licensed feeds and the sanctions lists, inside the desk's own tenant. We built that as a prototype with TMICC, a trader tested it for two hours, and it passed 7 of his 13 scenarios. This is what we learned.

Building a bilingual AI avatar for a UAE government library, native Arabic speakers caught three things our tests missed: the register, the pronunciation and the mouth. Formal Arabic with a voweled Emirati greeting, a speech-to-text check on every Arabic TTS phrase, and a geometric fix to the jaw took the demo to 20 correct answers out of 20.

Industrial land and warehouses sell on plot terms, not renders: setbacks, tenure, lease rate and distance to the port. These are the lessons from three plot explorers in Oman, Saudi Arabia and the UAE, including why massing must follow the setbacks and why warehouses lease by the unit.