NVIDIA Cosmos skills and tools sharpen ITMAX’s SamurAI platform, while City OS’s new AI agent enables city operators to simply ask
KUALA LUMPUR, 22 September 2026 – ITMAX System Berhad (“ITMAX” or the “Group”), an artificial intelligence-powered integrated digital infrastructure service provider, today showcased the latest generation of NVIDIA’s video-understanding and agentic AI technology, sharpening how Malaysian cities perceive, interpret and reason about activities across streets.
As Malaysian cities expand their camera and sensor networks, the opportunity is shifting from simply seeing more to understanding more, turning a growing stream of video into insights operators can act on immediately, and answers city teams can get without searching through dashboards. That shift, from detection to reasoning, is the focus of ITMAX’s latest work with NVIDIA.
The showcase builds on SamurAI, ITMAX’s computer vision platform, and City OS, ITMAX’s smart city operations platform, with an initial deployment now underway in Kuala Lumpur in collaboration with Kuala Lumpur City Hall (DBKL). The rollout is following a phased approach, starting with a first set of corridors and control-room workflows before widening across the city.
“Detecting something on camera is the easy part. Understanding what it actually means is what determines whether an operator can trust and act on it,” said William Tan, Chief Executive Officer of ITMAX. “NVIDIA’s stack has let us push SamurAI’s reasoning and accuracy forward in months rather than years.”
Building SamurAI V2 Faster on NVIDIA VSS
Development of SamurAI V2, ITMAX’s next-generation computer vision platform, accelerated this year on the back of the NVIDIA Metropolis blueprint for Video Search and Summarization (VSS). VSS’s modular microservice architecture and dedicated Smart City industry profile gave ITMAX’s engineers production-ready video intelligence and reasoning microservices to build on, rather than infrastructure to maintain themselves.
The result is a faster path from a new camera feed to a working, production-grade smart-city use case — a gain that matters as ITMAX phases SamurAI across Kuala Lumpur and, over time, to other Malaysian cities. Underneath, SamurAI’s video reasoning workloads run across NVIDIA H100, NVIDIA L20, and NVIDIA RTX PRO 6000 Blackwell Server Edition, letting ITMAX match compute to the deployment, from centralized data-center inference to on-premise city control rooms.
Cosmos 3 Brings Advanced Reasoning to SamurAI Inspector
The centerpiece of the showcase is the implementation of NVIDIA’s newly released NVIDIA Cosmos 3 world foundation model family into SamurAI Inspector, the SamurAI module responsible for turning raw computer-vision detections into verified, contextual alerts a city operator can act on.
Where earlier detection models could flag that something happened, a stopped vehicle, a person in a restricted area, smoke on a highway shoulder — Cosmos 3 gives SamurAI Inspector the ability to reason about a scene the way a human analyst would: weighing context, motion, and intent before classifying an event. Currently, Cosmos 3 is the leading open model on the VANTAGE-Bench benchmark for real-world, fixed-camera footage, directly relevant to the kind of road and public-space footage SamurAI processes every day.
The result is a marked step up in accuracy: in ITMAX’s initial Kuala Lumpur deployment, the new SamurAI Inspector has delivered more than a 70% improvement in alert verification accuracy, reducing false positives, and giving control-room operators alerts that reflect deep understanding of a scene and more confidence to act on them directly.
Teaching the Model Malaysia’s Roads
Off-the-shelf reasoning only goes so far, so ITMAX’s team fine-tuned Cosmos 3 using NVIDIA TAO fine-tuning microservices and built-in recipes for vision-language model post-training. Using supervised fine-tuning on Malaysia-specific road and incident footage, the team adapted the model to local conditions and edge cases, from accident detection to other roadway anomaly events, that a general-purpose model trained on global data tends to miss.
Similar results were shared in the NVIDIA Cosmos Cookbook recipe for Intelligent Transportation Post-Training, fine-tuning the Cosmos model on transportation-specific data measurably sharpens its accuracy on traffic-related reasoning tasks. ITMAX’s fine-tuned model shows the same pattern: markedly stronger reasoning on Malaysia-specific accident and road-anomaly scenarios than the out-of-the-box model.
SamurAI Shogun: Nemotron-Powered AI Agents
Reasoning and fine-tuning aren’t the only places NVIDIA’s models show up in SamurAI V2. SamurAI Shogun, ITMAX’s AI agent layer within the platform, runs on NVIDIA Nemotron open models, giving operations teams a conversational agent that can query SamurAI’s live data, walk through an event history, and trigger downstream workflows. The practical gain is speed: teams using SamurAI Shogun have cut the time needed to identify and search for a specific incident by more than 90%, turning what was a manual trawl through footage and logs into a single question. It brings the same agentic approach that powers City Maestro at the city level directly into the SamurAI platform itself.
From Dashboards to Dialogue: City OS and City Maestro
Alongside the SamurAI updates, ITMAX introduced City Maestro, an AI agent built into City OS, its smart-city operations platform, that lets city staff ask questions about city operations in plain language instead of navigating between dashboards. Under the hood, City Maestro orchestrates across City OS’s own services using the Model Context Protocol (MCP), pulling live answers from traffic, incident, and operational data rather than requiring an operator to know which system holds the answer.
“We didn’t want City OS to be another dashboard operators have to learn. City Maestro means anyone on the team can just ask — ‘how many incidents on this corridor today?’ — and get an answer immediately, without ever opening a report,” William added.
Together, the updates reflect a broader shift in ITMAX’s roadmap: from systems that only detect, toward systems that reason, verify, and converse in real-time. ITMAX said it intends to widen the Kuala Lumpur deployment progressively as each phase is validated with DBKL, and to extend SamurAI V2 and City OS to other Malaysian cities over time.
About ITMAX System Berhad
ITMAX System Berhad (“ITMAX” or the “Group”), incorporated in April 2001, is an artificial intelligence-powered integrated digital infrastructure service provider, playing a vital role in the establishment of smart cities. The Group is principally involved in the business of supply, installation and provision of public space networked systems, focusing on the supply and installation of networked lighting systems and networked traffic management systems, as well as provision of network facilities comprising video surveillance and communications network services. In addition, ITMAX also provides smart parking operator services to local councils. Leveraging its in-house research and development capabilities, ITMAX possesses the capacity to create various artificial intelligence features that are crucial in turning the smart city concept into a reality.
For more information, please visit https://itmax.com.my/
Media Query
Issued by Imej Jiwa Sdn Bhd on behalf of ITMAX System Berhad
Jason Chiew Tel: +60 17-980 4293 Email: jason@imejjiwa.com
Vince Teoh Tel: +60 16-441 9532 Email: vince@imejjiwa.com
Leave a Reply