Building with open-source AI models has a lot of benefits: it ensures privacy, gives the application owner control and transparency over the model lifecycle, and cuts costs at scale. In this talk, I will go through the state of open AI, workflows, tooling, and more for building with open models.

This presentation introduces Apertus, a fully transparent large language model initiative developed through collaboration between EPFL, ETH Zürich, and CSCS, representing a significant public-sector response to the concentration of AI development within private corporations. The talk examines its origin and fundamental challenges in current LLM deployment, such as opacity of training data, limited multilingual representation, and proprietary control. The initiative demonstrates how public research institutions can develop competitive language models while adhering to stringent ethical standards, including training exclusively on public data with copyright compliance and supporting over 1,000 languages.

As the organisations move past the initial excitement of Generative AI, they face a sobering reality: while building a single AI agent is easy, scaling agentic systems to deliver true enterprise impact is remarkably difficult. Currently, most organisations have moved fewer than 30% of their GenAI experiments into production. The primary hurdles? Siloed experiments, a fragmented LLM ecosystem, and a lack of inherent trust in autonomous decision-making. In this session, Serena Yuen, Strategic Sales Engineer at Dataiku, will discuss how organisations can transition from isolated AI concepts to robust, controlled agentic systems that deliver business value. Discover how a secure LLM Mesh and robust AgentOps practices can help you move from disparate POCs and cross the gap into full-scale enterprise AI transformation, along with all the principles data and architecture teams need to consider for true AI success. You’ll leave the presentation with a holistic understanding how AI at scale can deliver value with your people, orchestration and governance in mind.

As organizations evolve from digital to AI-native, data serves as the foundation for innovation, enabling AI to transform raw information into actionable insights. Cloudera powers modern AI-driven use cases, including Generative AI, by unlocking the full potential of enterprise data. This session will explore how Cloudera can help your organization accelerate AI deployment to production, without compromising performance, accuracy, or security.

AI agents offer new possibilities for automating work in enterprises, but their adoption is often limited by constraints around security, governance, system integration, reliability, and operational complexity. This talk discusses common challenges encountered when integrating AI agents into existing enterprise environments, shares practical lessons learned from real deployments, and includes a brief demo of the Unit8 GPT Wizard to illustrate how agentic workflows can be embedded into real systems.

AI rarely fails due to the algorithms, but because organization focuses on the technology instead of measurable outcomes. In this presentation, we share a value-first approach proven in the Swiss market that ensures to connect business priorities with strong technical delivery - so that AI creates measurable business impact rather than prototypes. Our method combines three elements that must work together: clear business goals, sector-specific insights, and pragmatic hands-on engineering. The journey starts with value discovery and how it will be translated in to clear KPIs and success criteria). We then demonstrate the impact through improvements in data, process, and systems. This builds the foundation required for reliability and sustainable adoption. Only this allows to introduce AI at scale for smarter automation, faster decisions, and consistent execution. The presentation concludes with real examples that illustrate how this method translates AI strategy into tangible business results, and what technical capabilities are required for an end-to-end delivery.

While recent studies suggest that 95% of AI pilots fail, we want to present a practical case study from the 5% that succeed. Lufthansa Group's Digital Hangar, in partnership with Cognizant Netcentric, successfully scaled SkAI—a production-grade GenAI Hub now deeply embedded across their internal data analytics ecosystem and beyond. Moving from initial prototype, to enterprise readiness required solving real challenges: governance to ensure compliant usage, seamless integration with existing enterprise systems, and adoption and training that drove 75% user uptake. The impact is tangible: 25% of users save 4–8 hours weekly on routine tasks. Attendees will gain actionable strategies for overcoming organizational barriers, fostering high-impact engagement measured through clear KPIs, and technical insights into how they successfully transitioned their agentic platform from proof-of-concept to sustained business value. This session will be presented by: Margot Löwenberg (Head of Data Analytics, Swiss International Air Lines) and Nicolas Athanasopoulos (Principal Data & AI, Cognizant Netcentric).

Following the rapid rise of autonomous AI agents in 2025, the enterprise security landscape is undergoing a critical transformation. While agentic AI offers massive productivity gains, it deepens the security risk through sophisticated attack vectors like memory poisoning and privilege abuse. At the same time, it also broadens the perimeter through Shadow AI - the unsanctioned use of AI-related tools by employees. This leads to a dangerous "visibility gap". Most enterprises lack the monitoring tools to detect unauthorized agentic workflows or local Large Language Model (LLM) deployments, leaving them exposed to data exfiltration and other attacks. The impending regulations around AI, such as the EU AI Act, are shifting liability to enterprises, making ignorance of Shadow AI a significantly costly legal and financial risk. To limit their exposure, enterprises need a strategic pivot from traditional block-and-deny tactics toward radical observability. Effective defense requires the automatic detection of unknown assets, cross-domain data correlation, and the rigorous management of non-human identities (NHIs). Security operations must evolve to employ agentic defenses - utilizing AI to monitor and counter other AI agents. To survive this "AI Wild West", companies must treat security as a business enabler rather than a barrier. Success in 2026 depends not on building higher walls, but on achieving total visibility and implementing automated, agent-driven response capabilities to secure the future of digital work.

Europe is no longer facing random cyber attacks. We are facing coordinated, state-sponsored operations now powered by AI. From ransoms to infrastructure sabotage, Advanced Persistent Threat (APT) groups are targeting enterprises, supply chains, defense contractors, and critical infrastructure with precision and speed. The reality stands that most organizations are already being targeted they just don’t see it. In this talk, we reveal what happens before an attack actually hits. We demonstrate how AI can detect early indicators of ransomware and advanced threats by learning attack patterns and correlating telemetry, predicting attacks before impact.

Generative AI has led to an explosion of content creation. Yet discovering relevant and inspiring content and products online is becoming increasingly difficult. Most digital experiences remain backward-looking, optimizing for historical interactions rather than understanding user interests in the moment. As a result, users are often trapped in narrow recommendation loops, while new and diverse content struggles to surface. This talk introduces a shift from AI generation to real-time perception: AI systems that continuously interpret user behavior and context as they evolve within a session. Real-time perception enables adaptive discovery, contextual search, and more effective agentic experiences; moving beyond static personalization toward moment-by-moment intelligence. Drawing from real-world deployments in large e-commerce and marketplace environments, the talk covers perception model architectures, product and system design, and concrete case studies with measurable business impact. Real-time discovery is emerging as a must-have capability for online platforms, one that directly translates into measurable gains in engagement and revenue.

Before LLMs, before dashboards and data warehouses, the 1980s had two disciplines quietly working on the right problem from opposite directions. Decision Support Systems asked: how can machines help humans make better decisions? Knowledge Engineering asked: how do we encode what experts know so that AI can reason over it reliably? Both asked the right question. Both fell short. The technology wasn't ready. 40 years later, we are living through an AI renaissance. Organizations are deploying AI agents that query databases, generate charts, and surface insights autonomously. And as organizations experience issues with using raw AI (wrong numbers, hard-to-verify answers, models that hallucinate with confidence) the lessons of the 1980s are coming back into focus. Reliable AI isn't about better prompts. It's about engineering knowledge. Key takeaways: - Why did these pioneering disciplines from the 1980s fail, and what has changed now? - What does Knowledge Engineering mean in the age of LLMs? - How does Knowledge Engineering turn Agentic Analytics from an "impressive demo" into a reliable system?

As industries such as transportation, energy, defense, construction, and space evolve toward increasingly complex, autonomous, and interconnected systems, the demand for robust safety and reliability measures continues to intensify. Prominent failures in recent years across various sectors have underscored the limitations of traditional, resource-intensive risk analysis methods and the urgent need for more effective solutions. Reshape Systems SA, a CERN spin-off, addresses these challenges with a proprietary, AI-driven SaaS platform purpose-built for comprehensive risk analysis across the entire project and product lifecycle. Leveraging advanced generative and explainable AI (XAI), our platform automates time-consuming hazard assessments, integrates supplier and design data, and ensures full transparency for expert oversight—meeting rising regulatory and customer demands for trustworthy, auditable AI. This automation enables early identification of potential failures, streamlines engineering processes, and delivers up to 80% savings in manpower, accelerating time-to-market while enhancing safety and reliability.

The talk outlines how generative AI enables Credura's intelligent insurance advisor and how AI impacts the software development space as a whole. The talk will showcase Credura's GenAI-powered product capabilities, which include Intelligent Document Understanding, AI-Generated Recommendation Narratives, and a client-centric approach to insurance advisory. S-PRO touches on the GenAI engineering backbone of such products and how software development in general is being impacted by Claude Code and other new GenAI tools.

We are drowning in AI demos that dazzle in the moment but fail in production. Why? Because Generative AI is incredibly good at returning an "echo" of your request - smooth and confident - but terrible at making a commitment. The biggest bottleneck for GenAI adoption in 2026 isn't intelligence or latency - it's trust. A real case of Oxagile’s 20+ years of engineering experience that converted into a trustworthy solution for a regulated Swiss organization (Pharma, FinTech).