Large language models can reason, generate code, and hold conversations. But they remain trapped behind screens. If AI is to become truly useful in our daily environments, it needs a body, a presence, and natural ways to interact with us.
In this talk, I’ll present Reachy Mini, an open and developer-friendly robot designed to explore what embodied AI can look like today. I’ll walk through how we are building its core software stack, from multimodal perception to real-time voice interaction, and why voice is emerging as the most natural interface for physical AI. We will look at how speech-to-speech pipelines, local inference, and modular backends allow Reachy Mini to move beyond scripted demos and into responsive, real-world interaction.
You will leave with a clearer understanding of what it takes to give AI a body, why voice-first interfaces matter, and how open tools can accelerate the next generation of interactive robotics.

"The model isn't sure", "The AI assistant wants to be helpful"…. We use these phrases daily as shorthand, but are they undermining your risk strategy? This talk argues that anthropomorphic language is a big sleeper risk in GenAI governance today. When we assign human verbs—thinking, deciding, lying—to probabilistic systems, we create an "Illusion of Intent." This linguistic drift isn't just a semantic annoyance; it is a governance hazard. It hacks human empathy, creates false trust, and obscures liability by treating system failures as character flaws. In this session, we will dismantle the habit of humanizing the machine. We will explore how precise, de-anthropomorphized language acts as a firewall for ethical safety and improves collaboration between Legal, Tech, and Product teams. Join us to learn why the most critical update to your governance framework isn't new code—it's a new vocabulary. Let’s stop governing the ghost and start governing the tool.

Problem:

Kaufmann argues that humanity stands at a turning point: machines and Generative AI will soon outperform us in many routine and even complex tasks, while our social, economic, and cultural systems are still built around compulsory work and industrial‑age roles of “Homo faber.” Without a new vision, fears of job loss, loss of control, and foreign dominance (China/USA) over AI systems will shape the future instead of our own European values.​

‍

Approach:

Kaufmann reframes robots and Generative AI as tools, like excavators or calculators, without intrinsic power fantasies, and insists we design them to augment humans rather than replace them. He sketches a near future where every person works with several humanoid or software agents that handle routine tasks, enabling humans to focus on uniquely human, hard‑to‑automate activities. He also calls for sovereign Swiss and European AI (e.g., SwissGPT, AlpineAI) to embed local values, privacy, and trust into foundational Generative AI infrastructure.​

‍

Key takeaways:

• Generative AI can trigger a shift from “Homo faber” to “Homo gaudens,” freeing people to pursue meaningful work and curiosity instead of mere survival.

• A “Eutopia” – a realistic, golden age – is possible if productivity gains from AI are used to reduce compulsory work and secure public finances.

• Trust, culture, and data protection will be the decisive “currency” in the global race for AI; Switzerland can lead by building reliable, privacy‑preserving Generative AI systems for governments, hospitals, and universities.

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.

For many financial institutions, Quality Assurance (QA) has ballooned into a massive liability, costing millions and stalling deployment. This session exposes the playbook leading banks use to turn this cost center into a competitive advantage. We explore the shift from manual grunt work to human-supervised AI agents. This approach delivers the speed of automation with the safety of expert oversight. We will analyze real-world cases showing how this specific AI implementation slashes overhead and dramatically.

Most companies are still experimenting with GenAI, but very few manage to scale beyond pilots—especially in highly regulated industries like insurance. In my talk, I will share how we moved from isolated GenAI use cases to designing and delivering the first Agentic AI–native claims-handling platform across three companies and five countries within a global insurance group. I will explain how Agentic AI changes process design, product architecture, and governance, and what it takes in reality to orchestrate multiple specialized AI agents while ensuring safety, reliability, and compliance with the AI Act, GDPR, and DORA. The talk focuses on practical learning: how to design agent roles and orchestration patterns, how to build trust and non-functional safeguards into autonomous flows, how to prepare the workforce for AI-centric operations, and how to balance speed with regulatory requirements. Attendees will leave with a concrete playbook for evolving from GenAI experiments to an AI-native operating model and with insights into the measurable impact on productivity, quality, and customer experience in claims handling.

Relying on black-box AI APIs often means trading data sovereignty and cost control for convenience. This session provides a practical blueprint for architecting a private, sovereign Model-as-a-Service platform using powerful open-source models. Attendees will leave with a concrete playbook to master observability, enforce security policies, and transition from an API consumer to a platform provider.

Building a generative voice assistant demo is easy. Getting it to production is hard. Scaling it to thousands of calls per day? That's where the real learning begins. This talk shares battle-tested lessons from taking a voice AI from prototype to handling thousands of calls daily - covering the unexpected challenges that no tutorial prepared us for.

Problem:‍

In sensitive application domains such as legal AI, impressive demos are easy to produce — reliable systems are not. What matters is the ability to systematically measure, compare, and control model performance. Without transparent statements about precision, error rates, and limitations, GenAI in real-world workflows remains opaque and risk-prone.

Approach:‍

The talk shows why high-stakes AI requires a clearly defined gold standard: structured data, annotated samples, and systematic benchmarking of model performance against human experts. With the emergence of agentic AI, many control steps can be significantly accelerated and partially automated as operational human intervention is reduced. The core principle remains unchanged: transparency about output quality. Model precision and error rates must be measurable and clearly communicated.

Key takeaways:

‍• Why demos fail as a proxy for real AI performance?
• Why human benchmarking remains essential even with agentic systems?
• How agentic AI increases speed without removing responsibility for output quality?

Imagine you're into playing instruments, singing, and producing music, but you’re lacking the decades of practice. Still, you write great lyrics and just "want to make music". As a personal project/side quest, I bridged that gap leveraging GenAI, my production skills, and professional post-production tools.

The result is Windlereye, a fictional alternative rock band with over 100 songs. Some of them are even good!

In this talk, I’ll demystify the "one-click" misconception by sharing details on my hybrid workflow and the workarounds I used to jump the biggest GenAI hurdles (vocal consistency, artifacts, instability). I'll explain how GenAI made me a better lyricist, and how I made my first whooping $0,000,001.12 in royalties.

Finally, I'll touch on my non-expert opinion on ethics and legal matters of this new frontier.

The future of decision-making will be won by organizations that deliberately pair human judgment with AI at scale: decision flows where machines grind through the data and people bring expertise, nuance, and accountability.

This session is for executives and senior leaders who don’t need to code models but do need to own the impact of AI‑infused decisions. We will unpack how data becomes wisdom through four layers, giving participants a clear, practical mental model for designing robust Human‑AI decision systems that are reliable, auditable, and safe to use in the boardroom.

Generative AI is beginning to reshape how news is created, translated, verified, and delivered — but for a global news organization like Reuters, innovation must go hand in hand with trust. In this talk, I will share how Reuters is approaching generative AI as both a powerful technological enabler and a responsibility‑critical capability.
Drawing on practical experience, the session explores how generative AI is being applied across the news lifecycle, including support for journalists and editors, workflow efficiency, multilingual content, and product experiences. The focus is not on experimentation for its own sake, but on real deployment decisions in a high‑stakes environment where accuracy, independence, and transparency are essential.
The talk will highlight key design choices and trade‑offs: where generative AI delivers clear value, where it must be constrained, and how human editorial judgment remains central. I will also discuss governance, risk management, and cultural adoption challenges when introducing generative AI into a trusted media organization.
The session concludes with practical lessons for media leaders and technologists navigating generative AI in environments where credibility is the product.

Enterprises have long relied on systems of record: databases and applications optimized for capturing and reconciling business data as the authoritative source of “what happened”. These human-centric architectures with rigid schemas and batch processes create barriers for agentic AI, which requires perceiving context, reasoning, and taking real-time action. This session explores the shift to systems of action, where an intelligence layer augments (rather than replaces) existing systems of record, enriching business objects with agent-generated insights to enable faster, smarter decisions.

GenAI promises to transform how organizations work — but turning that promise into secure, scalable, real‑world solutions is often harder than it seems. In this talk, discover how Migros built a flexible, enterprise‑grade chatbot platform that enables teams to create powerful assistants in just minutes, without compromising on security, governance, or quality.
Two years ago, we created our first chatbot — and quickly learned that technical innovation was only one part of the challenge. Alongside building early prototypes, we had to navigate aligning stakeholders and define governance models that satisfied both business needs and strict security requirements. This journey laid the foundation for what would eventually become our Chatbot Creator Platform.
In this session, we’ll show you that platform in action, share the lessons we learned while creating it, and explain how we built a system that now empowers teams across Migros to ship AI assistants in record time. You’ll walk away with practical insights for scaling GenAI in a complex enterprise environment — beyond the hype and toward real, sustainable impact.

Over 10,000 SAP BW systems face end-of-support by 2030 — a migration challenge so complex that traditional automation cannot scale to meet it. Manual migrations take two or more years, are error-prone, and require scarce expertise. We're using agentic AI to change that.

We'll show how multi-agent systems reverse-engineer legacy platforms, construct knowledge graphs of thousands of interdependent objects, and autonomously generate complete modernized systems — including the migration tooling itself.

Our three-phase framework — Reverse, Rethink, Rebuild — is a generalizable pattern for any complex system transformation, compressing years of work into months with automated validation at every step.

Attendees will leave with practical insights on multi-agent orchestration, knowledge graphs for spec-driven-development, synthetic data strategies for safe testing at scale, and hard-won lessons about where AI excels versus where human judgment remains essential.

Grounded in production experience, not theory.

Not a single day goes by without a new record investment in the field of artificial intelligence. The unit of measurement is no longer billions, but trillions and beyond. The arms race between tech giants seems limitless – and in any case defies the physical limits of our planet.

AI is infiltrating all areas of human activity at an unimaginable speed. Mastering artificial intelligence is the most significant quantum leap since the advent of the atomic bomb.

Does the future belong to the American and Chinese tech giants? Are Switzerland and Europe doomed to follow and submit?

There is an alternative. Sovereign, collaborative, efficient and ethical.

Giotto.ai seeks to exceed the current capabilities of AI to push the boundaries towards artificial intelligence capable of going beyond memorisation – towards reflection. Giotto.ai is developing technology that stands out for its ability to generalise tasks, solve problems and offer transformative potential for industries and society in general.

Way more efficient than large LLMs and infinitely less energy-and data-intensive, the solution advocated by Giotto.ai is based on two fundamental pillars: sovereignty and efficiency.

- Yes, it is possible to escape the frantic race for resources and infrastructure.
- Yes, it is possible to assert Swiss and European leadership in artificial intelligence that serves society, democracy and humanity.

Switzerland and Europe have all the talent, universities and centres of expertise needed to succeed. Together, let's create networks of investors, developers and conditions to control our own destiny.

Submitting is not an option. We can decide.

Biased GenAI products stifle innovation, from flawed data to homogenous teams missing real-world blind spots. How do you build diverse teams that catch biases early and deliver superior products? What rituals and skills ensure ethical, inclusive development from ideation to production? Join Priska Burkard to discover why diverse teams are your edge in GenAI success.

From generative AI transforming software development to advanced AI systems powering industry, defense, and public services, the mission for Europe is clear: AI sovereignty is no longer a political slogan, it is an economic, technological, and security imperative.

‍

Europe is making unprecedented investments in AI sovereignty – but most strategies stop at infrastructure. True sovereignty is a stack: from the data center and cloud environment, through model training and provenance, to fine-tuning, application orchestration, and day-to-day operational control. A gap at any layer means someone else holds the keys. This session unpacks the full sovereign AI stack, layer by layer, and shows what deliberate choices at each level look like in practice. We'll share real-world architecture patterns and outline what a mature, full-stack sovereign AI deployment looks like for regulated enterprises.

As foundation models commoditize, the real differentiator is how well you connect them to your own data and workflows. Avolta and D ONE share hard-won lessons from moving beyond pilots: consolidating data in a modern Lakehouse-as-a-Service architecture, enabling employees with internal copilots, and building focused business agents on proprietary data. We'll cover what we shipped, what surprised us, and three decisions we'd change if starting over today.

MCP adoption is accelerating, but many agent systems still misuse it. Instead of leveraging MCP as a runtime interface, tools are treated as prompt-time APIs: definitions are repeatedly injected into context, intermediate results are re-serialized, and valuable tokens are wasted. The result? Slower, less reliable agents, and the misleading conclusion that "MCP sucks, CLIs are great."

This talk challenges that narrative. We explore a better paradigm: code mode, where models generate small programs that call tools directly in a sandbox, dramatically reducing context overhead while improving multi-step accuracy. We’ll also cover dynamic tool discovery, ensuring only relevant capabilities are loaded when needed.

To make this practical, we introduce mcpc (https://github.com/apify/mcpc), an open-source universal CLI client for MCP. Built for real workflows, it combines code mode, persistent sessions, JSON outputs, and sandboxed execution, bringing MCP to where it shines: the shell.

In a live demo, we’ll orchestrate multiple MCP servers, transform results locally, and turn interactions into reusable scripts.

Hilda Liswani is a global executive with more than 15 years of experience advising leaders and building strategic partnerships across sustainability, technology, media, and finance. As CEO and Founder of WeBloom, she works with C-level decision-makers to navigate the forces reshaping business—AI, cultural change, leadership transformation—and to translate them into strategic clarity and organisational impact. Hilda specializes in creating high-trust, high-calibre environments where senior leaders can calibrate decisions, access cross-sector intelligence, and drive change at scale. Her work is grounded in a conviction that the future belongs to leaders who combine strategic foresight with human depth, and she has dedicated her career to equipping executives to lead with both.

‍

Generative AI is only as valuable as the structure behind it. While simple interfaces are great for prototyping, enterprise-grade autonomy requires a structured approach to ensure reliability and governance. In this workshop, we dive into the visual tools within Dataiku to build Autonomous Agents—AI workers designed to follow complex business rules, safely access enterprise data via the LLM Mesh, and execute multi-step tasks. We will explore the design patterns that separate "toy" use cases from sophisticated agentic loops, moving from basic interactions to robust engines that solve real organizational challenges.

Maximize your learning at GenAI Zürich with this hands-on workshop designed for executives, product managers, innovation managers and founders. In a condensed 50-minute sprint, your team will tackle a set scenario and build a minimal working prototype with the help of Claude Code. Immediate, expert coaching from S-PRO facilitators ensures rapid progress and technical breakthroughs. Perfect for maximizing hands-on experience and showcasing the speed of GenAI development. No prior coding or AI expertise needed!

Seamless communication is the key to unlocking global markets when scaling a business across borders. However, companies face a crucial dilemma. While general generative AI solutions offer broad capabilities, relying on them for enterprise localization introduces significant risks because these tools often produce generic text that dilutes brand nuance and mistranslates industry-specific terminology. This can result in a damaged brand reputation, financial losses, and legal and security risks.

In this fast-paced session, we will explore the critical differences between general-purpose GenAI and specialized language AI. We will present a targeted approach to global communication that prioritizes secure, purpose-built language models over generic tools.

In this session, you will learn how to:

-Ensure uncompromising quality: Use purpose-built language models to generate context-aware translations that protect your brand identity.
-Drive end-to-end efficiency: Deploy a centralized AI platform to speed up global operations and reduce costs securely.
-Connect the entire enterprise: Seamlessly integrate language AI across all business communications, from text APIs to real-time voice translation.

What if you could bridge the gap between a business idea and a working AI application in minutes without a massive engineering team? The way we build software has changed. Today, the most successful teams aren’t the ones with the most code; they are the ones who can turn data into insights the fastest.
In this interactive session, we strip away the complexity of traditional AI. We’ll show you how Snowflake Cortex acts as your "AI Co-pilot," allowing you to go from a simple question to a functional workspace using just natural language.

What you will experience:
- The "Zero-to-Hero" Workflow: See how a single prompt can automatically generate a full analysis notebook for you.
- AI for Everyone: Use the Snowflake CLI and Cortex to run powerful AI models on your data without writing complex algorithms.
- Speed as Your Superpower: Learn how to move 10x faster by letting the platform handle the heavy lifting, security, and infrastructure.
- The Future of "Prompt-Based" Building: A live demonstration of how software development is shifting from manual coding to intelligent orchestration.

Who should attend? Whether you are a curious lead, a data enthusiast, or a tech-savvy strategist, this session is designed to show you that the barrier to entry for Enterprise AI has finally vanished.

93% of developers say AI makes them more productive. Yet 91% of organizations report no impact on the top line. That gap isn't a model problem — it's a systems problem. Most teams are still treating AI like a magic autocomplete: powerful in the moment, inconsistent at scale. The result? Rework, unpredictable quality, and a growing trust gap between developers and their AI tools. In this talk, we'll explore four techniques that close the gap and turn unpredictable AI assistance into reliable engineering outcomes across the SDLC: Specs (defining what you want before the AI builds it), Steering (persistent behavioral guidance that shapes how AI operates), Skills (reusable task-specific workflows), and SOPs (step-by-step procedures that ensure consistency across complex tasks). Through real-world examples across coding, testing, and architecture review, you'll see why the teams getting the most from AI in their development lifecycle are the ones engineering the system around the model — not just the prompts they feed