K2K® eliminates data silos with dynamic intelligent knowledge map for city operators

Up to 256 IKM® Neural Networks per stream and sovereign IKM® Vision Language Models turn existing data into an intelligent and intuitive knowledge map, running on NVIDIA Cloud Partners and sovereign AI factories.


Nov. 3, 2026 | At Smart City Expo World Congress in Barcelona, K2K® demonstrates how IKM® Technology, built on IKM® Neural Networks and integrated with NVIDIA Cosmos 3 open world model and the NVIDIA Metropolis Blueprint for video search and summarization (VSS), extracts insights in real time from existing city and industrial camera networks.

K2K has expanded its agentic capabilities by introducing an intelligent knowledge map, connecting vision AI detections to city context so operators can understand what’s happening, assess its impact, and take action. Linking data across city assets, including roadways, intersections, waste collection points, vehicles, and operational process turns siloed detections into a connected operational picture.

Beneath every event is an architecture. Up to 256 IKM® Neural Networks work together to extract knowledge from every video stream, with an average processing time of 12 milliseconds, while IKM® Vision Language Models aggregate the scene, share semantic information and narrate it as it happens. Knowledge is built layer by layer, from frame to sequence, scene and narrative: every event is georeferenced and certified at source, and enters an intelligent knowledge map that inherits, evolves and compounds with every frame. Because each event is linked to the point and the instant in which it happened, the digital twin of a city or a plant is not a model to maintain: it is generated in wireframe, stays aligned with reality on its own and looks ahead, enabling predictive scenarios.


Built on NVIDIA Cosmos 3 and NVIDIA Metropolis

K2K uses NVIDIA Cosmos 3 to generate synthetic video of the rare events that matter most and post-trains the IKM® Vision Language Models based on the Cosmos 3 family. Cosmos 3 Nano and Cosmos 3 Super push accuracy even higher, while Cosmos 3 Edge, a compact 4-billion-parameter model, scales real-time analysis to thousands of cameras simultaneously. With NVIDIA VSS Blueprint integrated into IKM®, it enables video summarization and search capabilities that extend to live and recorded visual data.


From Smart Spaces to Industry

In Wroclaw, Poland, K2K is working to develop a shared AI infrastructure built with NVIDIA H100 GPUs, designed to support city services through a single accelerated computing environment. The initiative brings together K2K’s specialized neural networks, knowledge map, and vision language models (VLMs) accelerated by NVIDIA Cosmos and VSS Blueprint, to transform existing camera streams into actionable intelligence. The platform demonstrates how cities can unlock new operational capabilities from existing infrastructure without deploying separate AI systems for each service, providing a scalable architecture at any stage of deployment.

Across Smart Spaces, sensors that are already deployed across the territory become a network of active knowledge about the city. IKM® Neural Networks extract data to optimize mobility and road safety and to monitor the condition of roads and infrastructure, waste collection areas and the first signs of fires and flooding, while IKM® VLMs aggregate events and correlate them with weather and territorial data to reveal where congestion becomes structural and where critical situations tend to arise.

Every event takes its place on the digital twin of the city, generated in wireframe from the streams themselves, where it is seen in context and followed over time. Decision-makers ask in plain language, “What is the state of the territory right now?”, and receive a summary updated to the minute. Sovereignty, compliance and ethics are built in by design: people and plates are anonymized at source, data and inference stay within the territory that generates them, AI Services are designed in compliance with the GDPR and the EU AI Act, and decisions remain with people.


In Industry, IKM® Technology extracts the intelligent information of production processes, from refineries and port terminals to construction sites and production lines. Every event is certified at origin and becomes documentary evidence throughout the life cycle of the product, while process know-how never leaves the perimeter of the plant.


Sovereign AI Factories

Every connected camera creates continuous demand for computing, twenty-four hours a day. IKM® Platforms run on regional AI factories built on NVIDIA accelerated computing: as a service through NVIDIA Cloud Partners, or on sovereign on-premises infrastructure that K2K designs with cities, operators and industrial groups, with IKM® Neural Networks are also available pre-installed on servers with NVIDIA GPUs. Inference stays inside the perimeter that generates the data: every site is a node, knowledge travels between nodes and data stays where it was born. K2K creates and sustains the demand for computing and scales through an ecosystem of technology partners, system integrators and local partners.


###


About K2K®

K2K® produces complex Neural Network Architectures that extract, analyze, synthesize, manage and share knowledge in real time, based on Video AI, Video GenAI, VLMs and Video Foundation Models. Its proprietary Intelligent Knowledge Management® (IKM®) technology understands the meaning of video streams and correlates it into structured, certified knowledge for smart spaces, ports, industry, energy and critical infrastructure. www.k2k.ai

Media contact: K2K® | k2k@k2k.ai