Industry Session
Implementing Secure AI Agent Use Cases in Enterprise Environments
Mira El Kamali, Elmira Gazizova, Eric Gagnaux — Key-IT
Small and medium-sized enterprises (SMEs) increasingly accumulate large volumes of digital information, yet much of this knowledge remains underused. Content is dispersed across tools and platforms, and many workflows are still manual and time-consuming. While generative AI technologies have become more accessible, organisations often struggle to move from experimentation to secure, reliable, production-ready solutions that integrate with existing systems, governance frameworks, and compliance requirements. This presentation focuses on the secure integration of AI agents in enterprise environments. It presents practical scenarios in which AI agents are embedded into daily workflows to improve productivity, streamline communication, and orchestrate internal automation, while respecting constraints such as data governance, identity management, and auditability. The implementations build on Microsoft Copilot, Azure AI services, and workflow orchestration tools such as n8n to deliver solutions that are adaptable, maintainable, and aligned with enterprise security standards. The first category of use cases demonstrates secure automation of communication and repetitive tasks. These include generating context-aware emails for large recipient groups using validated external data, producing templated documents for specific teams while ensuring content integrity, and retrieving and synthesising information stored on secure platforms such as SharePoint, in compliance with access permissions. These workflows show how AI agents can operate within existing identity and access controls, reducing manual effort without compromising transparency or user oversight. The second category presents an internal AI automation pipeline designed to enhance document understanding and reuse. Enterprise documents are automatically collected and summarised at multiple levels, from detailed summaries to higher-level overviews. All outputs are fully traceable, allowing employees to inspect results, manage prompts, and link each summary back to its original source. The presentation discusses architectural and operational design choices for deploying AI agents at scale, with a strong emphasis on security, access control, and human oversight. Results are demonstrated through operational dashboards and interfaces that make AI outputs visible, auditable, and reusable. Overall, the talk provides practical insight into how organisations can deploy AI agents in a secure, transparent and enterprise-ready manner.
Transforming Industrial Operations with Multimodal AI
Fabien Roth, Joachim Ott — Philico SA
Manufacturing environments depend on fast, accurate access to technical knowledge, yet this knowledge is typically fragmented across multilingual and multimodal sources such as operational procedures, standards, schematics, technical diagrams, and visual inspection data. Engineers and shop-floor operators are expected to make precision-critical decisions under time pressure, while today's off-the-shelf AI assistants fall short of industrial requirements. They struggle with exact reasoning, structured information extraction, and the reliable use of visual and graphical content, often producing incomplete or even fabricated outputs that are unacceptable in safety- and compliance-critical contexts. In this talk, we present how multimodal AI can be engineered to meet industrial standards of precision, trustworthiness, and operational value. Drawing from real-world insights from industry partners, we introduce an AI system that combines multilingual and multimodal information extraction with controlled, context-aware generative reasoning. By structuring knowledge across text, tables, diagrams, and images, and by tightly coupling retrieval and generation, the system ensures that responses are grounded in validated, relevant sources. Attendees will gain practical insights into why multimodal AI is essential for industrial operations and how moving beyond generic AI unlocks efficiency gains and safer, more reliable decision-making.
Intégration de l’IA dans les Services d’Information Géographique (SIG): de la GéoAI aux assistants IA sur ArcGIS
Maria Algora Tussy — Esri Suisse
L’objectif de cette présentation est de montrer comment l’IA intégrée dans ArcGIS permet d’optimiser les analyses, d’automatiser des tâches complexes et de rendre les SIG plus accessibles à tous. Nous illustrerons d’abord comment les dégâts agricoles peuvent être évalués après une tempête grâce à des modèles pré-entraînés guidés par langage naturel (GeoAI), résultant dans une cartographie des surfaces agricoles détruites accompagnée d’indicateurs associés. Nous verrons ensuite comment analyser des accidents de circulation à l’aide de scripts ArcPy générés avec l’assistant IA, permettant de produire une nouvelle classe d’entités indiquant le nombre d’accidents par route. Un autre cas d’usage portera sur Survey123 et son intégration de l’IA qui permet d’analyser automatiquement des images et de renseigner les champs d’un formulaire en fonction des résultats obtenus. Finalement, nous présenterons le Data Explorer, un agent autonome capable d’exécuter des analyses spatiales et statistiques et de créer des nouvelles couches de données. Cet outil facilite l’exploration des géodonnées pour les utilisateurs non spécialisés dans le domaine.