Open Source Summit Europe 2024

We will explore Conversational AI that works like the Web. In this talk, we explore the pivotal role of OVON Specs in the Open Voice Network's Interoperability Initiative. We highlight how OVON Specs are crucial in crafting a universal, open-source AI voice interface for seamless cross-platform assistant integration. We'll delve into OVON Specs' practical applications and collaborative potential, showcasing their versatility and impact in the evolving landscape of conversational AI, and how they're shaping the future of voice technology. When we talk about the combination of Voice and AI, it’s not just about a new way of communicating, instead, we are dealing with a human biometric feature unique to us. It's useful to design standardization protocols and guidelines for Conversational AI interoperability and scalability: we will run through the current architecture proposed by the Open Voice Interoperability group, part of the Linux Foundation, working on the Conversational AI Universal API Message envelope specifications for interoperability: we will share how to play with the available sandbox and some ethical use cases powered with AI Retrieval Augmented Generation as well.

Demand for generative AI development is moving like a bullet train! Rapid innovation can move lightning-fast and produce exciting projects. Now is the time to embrace open development, refine best practices, and collaborate on standards for all to benefit. This session will outline common pain points involved in building LLM-based generative AI applications, especially those using RAG techniques, and connect them to open solutions. We will share reference architectures to help shorten developers’ paths to releasing performant AI applications to meet the needs of stakeholders and users. Finally, we’ll share community efforts, such as the Linux Foundation’s Open Platform for Enterprise AI project, to advance this critical work. Join us to explore ways to discover the untapped potential in generative AI development workflows.

Milvus, an LF AI project, is an open-source vector database built to power Gen AI solutions. 80% of the data in the world is unstructured data, and vector databases are the databases that help you get valuable insights from unstructured data. With this in mind, we built Milvus as a distributed system on top of other open-source solutions, including MinIO and Kafka, to support vector collections that exceed billion-scale. This session will deeply dive into the architecture decisions that make this cloud-native vector database seamlessly scale horizontally, provide users with tunable consistency, orchestrate in-memory and on-disk indexing, and scalable search strategies.

Several open models of varying size, quality, and performance have been released over the past twelve months. We have recently announced InstructLab, a project that allows contribution of skills to an existing model without the need to fully fork and fine-tune. This project establishes an upstream community built on contributing acceptance workflows for models. This exciting technology aims to make open source AI more approachable. In this session we'll explain how communities and individuals can contribute domain knowledge to models incrementally, in a unified and open way, while reducing model variations and resulting in an improved version by augmenting what’s already there.

Strategic position of AI development in industrial settings, the surrounding regulations, and organizational challenges, have an interesting analogy in how we deal with Open Source Software on industrial scale. Definition of governance as set of processes and guidelines for decision making within an organization gives us an opportunity to apply similar abstract models to seemingly different areas of endeavor such as AI production and consumer organizations, on one side, and Open Source Software, on the other. The learnings from running technical operations of OSPOs translate remarkably well into this model. Intake, compliance, distribution, so well known in the world of Open Source Program Offices, with its federated democratized governance, are some of the pillars of the processes that drive AI production and consumption organizations. In this talk we will try to shed some light on how in Volvo Cars we approach practical aspects of these growing processes.

What if you could harness the power of artificial intelligence to make it easier to employ design thinking principles to benefit the users of your application? Design thinking is a highly effective process in designing successful applications. Design thinking can help application developers identify the problems that the application is solving and can improve user experience at launch. But to be most effective design thinking must be employed straight from concept to release. However, this takes money and time which start-ups, or small companies may not have. In this talk we provide an overview of a design thinking system of open source Large Language Model (LLM) agents to help developers get the most out of the design thinking process. Each agent is primed with a persona of an individual in the process and will work with the development team to produce design thinking artefacts and deliverables. We will show how the system helps to accelerate both the effect and adoption of design thinking in an organisation. This is the talk for you if you want to understand how Generative AI can provide solutions that traditional programming may not.

Bias in AI has become a hot topic and increasingly, we are seeing how dangerous it is. Dangerous because AI is gain more influence in decisions that impact lives, decisions about who gets employment, healthcare or economic development and who gets profiled for crime, extra search at the airport, and so on. Even tools designed to identify AI-generated content have shown bias against non-native speakers based on their choice of verbiage. To reduce bias, experts often refer to having diverse training data and cognitive diversity, but where else is this achievable than in Open-source communities? This session is designed to open up discussions around reducing bias in AI with Open source. Abubakar will start by sharing examples of biases, sharing strategies to leverage open-source communities, and opening the floor for attendees to share their opinions and views, with the goal of creating a resource that will be valuable to the community.