From AI “agents” to generated simulations and large language models, AI is revolutionizing how we see, measure, and understand the world around us. This session showcases cutting-edge approaches to geospatial intelligence, from automating road and asset detection with minimal manual input, to turning large-scale environmental and infrastructure changes into actionable insights, and creating fully navigable indoor maps for massive facilities in minutes. Together, these presentations highlight how AI – through training models, agentic systems, and generative techniques – can improve accuracy, reduce human effort, enhance scalability, and make advanced geospatial analysis accessible to organizations of all sizes. Attendees will gain a peek into the future at some of these cutting-edge technological advancements, and emerging opportunities for leveraging AI to solve real-world geospatial challenges.
The following presentations will be shared in this session:
Developing AI Models using Low Volume Training Data: Making AI Asset Detection Accessible to Small Entities
Presented by Suliman Gargoum, Reflektar
The use of geospatial remote sensing technology, such as lidar and cameras for asset detection and inventory has gained momentum in recent years. However, the data is often manually reviewed to extract asset information, which can be time consuming and inefficient. As a result, many agencies are working on automating the process by leveraging artificial intelligence (AI) including supervised deep learning technology. Despite the huge potential and the high accuracy achieved in previous work, one drawback of supervised deep learning models for road object detection is that they require a large human labelled dataset for training. This often involves multiple trained individuals labelling data over hundreds if not thousands of kilometres, which is something some small agencies cannot afford. Supervised deep learning models also often fail to generalize when tested on out-of-domain or unseen objects. To overcome these issues, this session presents a novel few-shot learning approach to automate road feature detection from georeferenced panoramic images, without the need for manual labelling of a large amounts of data. The proposed approach enables models to generate a large synthetic, yet realistic, automatically labelled dataset that can then be used to train the deep learning models. The approach is designed to generalize effectively using a very limited set of labelled instances achieving comparable accuracy levels to models trained using large datasets. This allows the model to be easily re-trained for different scenarios and for different assets unlocking the true potential of AI technology for object detection. The presentation also demonstrates the success of the proposed approach on data from various cities and diverse road environments.
The AI Mapping Benchmark: How AI is Transforming Indoor Mapping at Scale
Presented by Ege Akpinar, Pointr
MapScale® is Pointr’s AI-powered mapping engine designed to create, update, and maintain indoor maps at enterprise scale. Built to support billions of square feet across smart buildings, airports, hospitals, events, and shopping malls, MapScale turns outdated floorplans into fully navigable digital maps in minutes no manual labor required. With advanced geometry recognition, POI tagging, and scalable automation, MapScale is already transforming how the world’s biggest organizations manage their indoor spaces.
But how well do today’s leading AI models perform when given the same task? To find out, Pointr created the first-ever AI Mapping Benchmark, challenging GPT-4o, Gemma 2, and others to generate indoor maps from raw floorplans. The results reveal not only the current strengths and limitations of general AI models in geospatial tasks but also why purpose-built engines like MapScale are leading the way.
This session will take you through both MapScale’s core technology and the benchmark findings, offering an honest, data-driven look at where AI mapping stands today and where it’s headed.
Breaking the Code Barrier to Unlock Earth Observation Data with Conversational AI.
Presented by Brent Mitchell, RedCastle Resources, Inc.
Earth observation data, while increasingly vital for monitoring our planet, remains difficult to access and analyze for many potential users. The complexity of traditional platforms and specialized workflows creates a barrier to entry, hindering rapid decision-making and broad application. This presentation introduces ASKTERRA, a novel platform designed to bridge this gap by leveraging a simple, conversational interface to provide access to powerful geospatial analysis capabilities.
The platform utilizes Landsat and Sentinel-2 satellite imagery to conduct near real-time change detection on a global scale. In addition, we have incorporated Google DeepMinds’s WeatherNext models that enable the prediction of climate variables and associated risk models 10 days into the future. By integrating expert-developed models and analysis workflows into a user-friendly chat interface, the platform allows users to explore and visualize global, near real-time changes and risks without the need for extensive technical expertise. This results in more confident geospatial users and accelerates their timeline for identifying environmental changes, improving landscape knowledge, and supporting more informed decision-making across various fields.
Turning Change Into Insight: Advancing Geospatial Monitoring with AI and Agentic Systems
Presented by Woolpert
Monitoring change at large scale and high resolution is becoming essential across sectors—from climate risk assessment and land use management to infrastructure oversight and economic monitoring. With more frequent and granular data from satellite and aerial imagery, the challenge is no longer just detecting change—it’s interpreting it quickly and effectively.
Deep learning has made it possible to automate many aspects of geospatial analysis, identifying where and when meaningful change occurs. But translating that information into decisions often still requires expert interpretation, complex workflows, and specialized tools.
This is where Agentic AI adds a critical layer. By combining domain-specific models with natural language interfaces and reasoning capabilities, agentic systems can help users interact with complex geospatial data more intuitively. Whether it’s an analyst seeking early signs of environmental stress, or a policymaker exploring infrastructure vulnerabilities, Agentic AI systems can surface relevant insights, answer contextual questions, and guide users through decision-making processes—regardless of their technical background.
We’ll highlight examples built using the Google Cloud stack, and walk through real-world use cases in climate risk, resource management, and infrastructure oversight—showing how these technologies can enhance both situational awareness and responsiveness.
