"KI ist gekommen, um zu bleiben", ist von vielen Seiten zu hören. Das stimmt. KI ist nicht mehr wegzudenken. Millionenfach finden wir KI in Anwendungen und Funktionalitäten wieder. Die Entwicklung ist rasant und viele wollen noch "auf den Zug aufspringen". Darunter sind auch Lösungen, die ohne KI auskommen und sogar effizienter wären. Umso wichtiger ist es, bei der Planung zu erkennen, wann ein Problem auch ohne KI gelöst werden kann. In dieser Session lernen wir ein paar komplexe Beispiele aus der realen Welt kennen und zeigen auf, wo KI nützlich ist und vor allem wo nicht.

Zielpublikum: Projektleiter, Produktmanager, Anfänger in der Entwicklung
Voraussetzungen: Keine
Schwierigkeitsgrad: Advanced

Generative AI is the talk of the town. Anyone who spends just five minutes thinking about AI can surely come up with several useful business use cases. However, all too often, we find ourselves facing the following dilemma: we want to quickly launch our chatbots and assistant systems and bring our ideas to market readiness. Yet at the same time, important, complex, cross-functional aspects such as data protection, compliance, operational readiness, or model fine-tuning often slow down rapid development and deployment.

Furthermore, enterprise scale AI projects often involve many different stakeholders: data engineers, AI specialists, software engineers, operational experts, and business departments. Too much talking and no progress at all are the result.

AI platforms to the rescue! We believe that established platform engineering approaches and technologies, combined with LLM Ops practices, can tackle this dilemma. Only a robust, scalable, and flexible platform enables our teams to efficiently develop, operate, and manage their data, models, and applications. The platform hides the inherent technical complexity, while allowing users to fully focus on the use case and the creation of value and innovation.

We will explore what a corporate AI platform can look like and the components and services it requires. We discuss how a company-wide platform strategy not only simplifies technical implementation but also creates an ecosystem for innovation, fosters collaboration, increases reusability, and ultimately drastically shortens the time to market.

Target Audience: Developers, Architects, Data Scientists, Decision Makers, Platform Engineers
Prerequisites: Basic knowledge in AI
Level: Advanced

The EU AI Act was established to promote trustworthy AI. This legislative framework outlines comprehensive and rigorous requirements for AI systems, sorting them into four risk groups: prohibited, high-risk, limited-risk, and minimal-risk AI systems. With a grace period of 6 to 24 months from the regulation's publication, it is recommended for companies using AI to start preparing for the upcoming requirements. To ensure technical readiness for compliance with the AI Act, it is crucial to integrate established data governance, AI governance, and MLOps techniques throughout the life cycle of an ML/AI system.

My talk is intended to provide practitioners with a thorough understanding of the EU AI Act's requirements. They also offer practical guidance on implementing MLOps, observability, and data governance processes that align with the regulation. While exploring the legislation to identify the requirements for AI systems and general-purpose AI applications, our discussions will provide an engineering perspective on the proactive implementation of the provisions of the EU AI Act.

Target Audience: Architects, Developers, ML Engineers
Prerequisites: MLOps
Level: Advanced

GenAI services have been rapidly integrated into various digital business models, but what if your data holds better answers? How can this technology be combined with an organization's knowledge and data?

This talk explores Large Language Models (LLMs) and their augmentation with custom data via Retrieval-Augmented Generation (RAG). Discover Statista's pioneering journey from rich search results to concise, informed answers with their LLM-based application, ResearchAI. We'll discuss challenges such as building a skilled team, the impact of exclusive data on answer quality, high costs, query latency, and LLM hallucinations despite accurate data. This session provides a realistic look at the hurdles and strategies for optimizing RAG applications in the real world.

Target Audience: Data Scientists, Data Engineers, Developers, Decision Makers
Prerequisites: Basic knowledge of GenAI
Level: Expert

Data Contracts sind so etwas wie APIs, aber die Datenwelt funktioniert doch etwas anders. Ein Data Contract definiert das Datenmodell der bereitgestellten Daten und deren Qualitätsattribute in einem YAML-Format. Data Contracts enthalten zudem semantische Definitionen, SLAs, und legen die Nutzungsbedingungen für die Verwendung von Daten fest.

Data Contracts sind in erster Linie auch ein Kommunikationsinstrument, um ein gemeinsames Verständnis darüber auszudrücken, wie Daten strukturiert und interpretiert werden sollten. Sie machen implizite semantische und qualitative Erwartungen explizit. Später in der Entwicklung und Produktion dienen sie auch als Grundlage für die Codegenerierung und automatisierte Tests, um sicherzustellen, dass die bereitgestellten Daten auch der Spezifikation entsprechen.

In diesem Vortrag möchte ich die Data Contract Specification (datacontract.com) und das Data Contract CLI zur Validierung von Data Contracts in CI/CD-Pipelines vorstellen.

Zielpublikum: Data Engineers, Softwarearchitekten, IT-Verantwortliche
Voraussetzungen: Verständnis über grundlegende Datenarchitekturen
Schwierigkeitsgrad: Basic