Presentations
Discover how to bring Entra ID authentication on Ubuntu Desktop at scale with Landscape, Authd, and cloud-init. Join Ken VanDine and Rajan Patel.
Euphony Echo is a VR app making music therapy accessible for individuals with autism, supporting emotional expression, anxiety management, and communication. Users explore songs backstage, perform on a safe virtual stage, and receive real-time pitch feedback via an advanced Vocal Engine. By blending technology and therapy, Euphony Echo fosters confidence, reduces anxiety, and bridges gaps in traditional music therapy access.
With the rise of large language models (LLMs) enhanced by retrieval augmented generation (RAG), it has become essential to develop rigorous evaluation methodologies to assess their effectiveness across diverse use cases. RAG combines a model's generative capabilities with information retrieval, allowing for contextually relevant responses grounded in up-to-date, factual knowledge. This talk will focus on the unique challenges and best practices for evaluating RAG applications covering quantitative metrics (e.g., accuracy, relevance, etc).
Nix evaluation performance is a known, long-standing issue to the community. This talk will cover a benchmarking setup, concessions to that setup made to retain the author's sanity, and ways to improve evaluation performance and their trade-offs.
I worked at GoTo.com/Overture/Yahoo in Pasadena 24 years ago, we were doing Big Data when it was called just data. Information Retrieval used all the sexy ML/LLM/AI concepts you hear about now. I'll discuss the parallels and extrapolate on what's next for LLM, based on what happened with Search.
On its face, bpftrace is a simple DSL/tracing-tool for writing BPF programs. It abstracts away a lot of the complicated user and kernel space code neccessary to write observability and debugging programs. However, adding new language features/syntax isn't easy. There are a lot of considerations to be made in regards to bpftrace's primary usecases and the classic problem of how much complexity/details to hide from the user.
We migrated our Slurm based AI research infrastructure to Kubernetes, focusing on a seamless user experience. This required innovative solutions for managing complex systems, including provisioning, authentication, access control, and host management.
We leveraged Kubernetes features to build a scalable and flexible infrastructure, and developed new approaches to configuration management. Join us to learn about our journey and the lessons we learned along the way.
We are used to calling Metrics, Logs, Events, Traces as the 4 pillars of Observability. True Observability is not attainable without being able to jump from one pillar to another. Exemplars promote the same from Metrics to Traces. Over the years, we have found extreme value in being able to use Exemplars for other things such as making tail sampling decisions, enriching alerts with sample traces and finally serving longer term Exemplars out of our tracestore built on top of ClickHouse. In this talk, we discuss how metrics and traces go hand in hand through the concept of Exemplars.
This talk will mention about the world of extensions of PostgreSQL.
Learn about the ecosystems in the Fedora and CentOS projects, and how you can package software for both. We will present an overview of the Fedora project followed by an overview CentOS ecosystem. We'll show how both systems are developed and what the various special interest groups are doing. We will then provide a packaging workshop where you can learn how to package your favorite software, whether you want it in Fedora, EPEL, a CentOS SIG, or just for your own personal use.