Industry Demos
Industry Demo #1
Title: SOMIN: An Explainable AI and LLM Platform for Real-Time, Data-Driven Digital Marketing Strategy
Authors: Aleks Farseev (SOMIN.AI, Singapore)
Presenter: Aleks Farseev (SOMIN.AI, Singapore)
Abstract: The SOMIN platform demonstrates how explainable AI and large language models (LLMs) can transform digital marketing. The system offers a full-stack solution for competitor ad analysis, content scoring, persona generation, and AI-driven campaign storytelling. At its core, a predictive model ranks ad content based on an expected click-through rate, with attention-based heatmaps providing visual explanations for the scores. An LLM-powered pipeline analyzes competitor ads to extract strategic insights and identify customer personas, each with a generated narrative to guide campaigns. The live demonstration is interactive, allowing users to filter content and trigger the generation of campaign stories in real-time. Deployed at multiple marketing agencies, the platform has already yielded a significant 2x to 12x improvement in planning efficiency, allowing teams to focus more on creative strategy. This demonstration is ideal for researchers and practitioners interested in explainable AI and intelligent content systems. The full demo video is available at https://youtu.be/8DlKHAFf6qI.
Industry Demo #2
Title: A Streamlined System for Multimodal Industrial Anomaly Detection via 2D and 3D Feature Fusion
Authors: Wenbing Zhu (Rongcheer Co., Ltd., China), Bo Peng (Shanghai Ocean University, China), Mingmin Chi (Fudan University)
Presenter: Wenbing Zhu (Rongcheer Co., Ltd., China)
Abstract: We demonstrate an end-to-end system for real-time, multimodal industrial anomaly detection (IAD), built upon a custom hardware platform for synchronized 2D and 3D data acquisition. Our core contribution is a novel cross-modal residual mechanism that identifies defects by quantifying predictive errors between visual and geometric feature spaces. Instead of traditional concatenation, our dual-stream architecture mutually predicts features across modalities, leveraging the prediction residual’s magnitude as a direct and robust anomaly indicator. The entire system achieves sub-second inference from acquisition to decision, enabled by efficient depth map analysis that circumvents the complexity of direct point cloud processing, offering a deployable solution for high-speed inspection.
Industry Demo #3
Title: IDPFlow: A No-Code Agentic Framework for Multimodal Intelligent Document Processing
Authors: Goutham Vignesh, Harikrishnan P M, Siddartha Reddy, Saisubramaniam Gopalakrishnan (Phi Labs, Quantiphi India)
Presenter: Goutham Vignesh (Phi Labs, Quantiphi India)
Abstract: Intelligent Document Processing (IDP) is critical for unlocking insights from documents, but traditional systems are hindered by their reliance on specialized data science teams. This paper introduces IDPFLow, a novel framework that empowers domain experts—such as business analysts—to independently build and deploy sophisticated IDP workflows without writing any code. IDPFLow uses a powerful agentic architecture with a versatile toolkit to execute high-precision tasks like document classification, visual question answering, and information extraction. Its dynamic workflow proactively suggests a comprehensive data schema upon document upload, transforming the user’s role from builder to supervisor. This initial workflow can be refined in real-time through simple conversational instructions. The system’s intelligence also allows it to adapt existing workflows for new tasks, dramatically reducing development time. For complex or long documents, it uses a Multimodal Retrieval-Augmented Generation (MMRAG) pipeline to overcome context limitations, ensuring robust information retrieval from both text and images. To foster user trust, IDPFLow includes a grounded traceback citation mechanism that highlights the exact document segments from which the agent derived its responses. This framework is projected to reduce workflow creation time by 60-70%, bridging the gap between technical complexity and domain expertise.
Industry Demo #4
Title: XReco Platform and RAI News Media Demonstrator
Authors: Roberto Iacoviello (Radiotelevisione Italiana, Italy)
Presenter: Roberto Iacoviello (Radiotelevisione Italiana, Italy)
Abstract: The Horizon Europe XReco project presents a data-driven ecosystem for next-generation media production, focusing on extended reality (XR) and virtual production. The platform unifies the ingestion of diverse media types—including text, audio, video, and 3D assets—to overcome archive fragmentation and enable efficient content access. A core feature is the Orchestrator module, which provides powerful cross-modal semantic search, allowing users to find content across different formats using neural descriptors. The platform also includes tools for enhancing media quality and generating 3D models from 2D assets using state-of-the-art techniques like Gaussian Splatting and NeRF. For a TV documentary on Guglielmo Marconi, XReco was used to create detailed 3D models and integrate digital avatars and holoported experts into a virtual museum. By supporting integration with third-party tools like Unreal Engine and Unity, XReco streamlines modern XR workflows. The platform is a robust end-to-end solution for the technical and commercial complexities of virtual production, paving the way for innovative storytelling. The demonstration will feature a walkthrough of these functionalities, a viewing of the documentary, and a hands-on session for attendees. The video is available at
https://drive.google.com/drive/folders/15XTkg-x1U62hQ2dRo2ABiT3LG94CJXO5.
Industry Demo #5
Title: Real-time GenAI solutions for video streaming in low-bandwidth settings
Authors: Matteo Bruni, Claudio Baecchi, Fabio Clabot (Small Pixels, Italy), Marco Bertini (Small Pixels / Università di Firenze, Italy)
Presenter: Marco Bertini (Small Pixels / Università di Firenze, Italy)
Abstract: Small Pixels, a University of Florence spin-off, has developed a GenAI-powered solution for real-time video compression in low-bandwidth environments, specifically for surveillance applications like bodycams and drones. Traditional video compression often results in low-quality, artifact-ridden streams that hinder the identification of people and vehicles. This issue is exacerbated by high movement, common in security footage. To solve this, Small Pixels’ technology either processes video at the edge to make it more compression-resilient or captures low-resolution video and reconstructs it to Full HD using real-time GenAI upscaling. The system’s models are trained to optimize images for compression, resulting in sharper details and up to a 50% reduction in bandwidth without requiring camera upgrades. The solution enhances object detection accuracy by up to 20% and perceived video quality by 5%. It delivers Full HD at 700 kbps, saving 30% on bandwidth. This technology is crucial for law enforcement and security, ensuring critical visual details like faces are preserved in dynamic footage. Validated through a public safety program in Singapore, the system is easy to deploy and maintains continuous, clear surveillance in unreliable network environments. The live demonstration will showcase the technology’s performance with police bodycams.