Presentations

With the rise in popularity of AI applications, enterprises dive headfirst into development without having the proper foundations in place. AI workloads require heavy resource usage, and few enterprises lack robust infrastructure to handle them efficiently. These AI workloads often involve sensitive data, large-scale data movement, and high-performance compute nodes that require secure communication between components. Typical network security in Kubernetes is no longer limited to isolating services. It now includes protecting model training pipelines, securing inter-node traffic, and enforcing policies that ensure data confidentiality and compliance. The most common challenges enterprises face while developing AI applications are overprovisioning resources, an old-school infrastructure setup (a VM-only mindset), and insufficient security to prevent cybersecurity risks and protect their data. 

In this session, we’ll walk through best practices for building optimal AI infrastructure while utilizing k0rdent and Kubernetes, and also leveraging Cilium to maintain stringent data security and compliance.

We’ll cover:

• Why over-provisioning resources is such a common error with AI infrastructure, and best practices for efficient use of resources for maximum ROI.
• How to move beyond a VM-only mindset to a more modern, Kubernetes/bare-metal–aware platform that can keep up with AI teams’ needs.
• How to safeguard AI workloads without sacrificing the scalability that makes Kubernetes effective in the first place.

With proper, strong infrastructure, AI workloads will run smoothly, securely, and without the usual operational overhead. Whether you’re a platform engineer, an AIOps Engineer or someone who wants to get into AIOps, you’ll benefit from this talk on best practices for creating optimal and secure AI infrastructure. 

It's 2026. Installing Linux, spinning up a VM, and launching containers is easy. So why does building a real cloud at home and in enterprise still feel like an enless IKEA maze? And where are the do-it-yourself instructions? Drawing on a decade of experience deploying to homelabs, enterprise clouds, even submarines, and satellites, we'll talk the talk and walk the walk live. Together, we'll build your own cloud, command by command and click by click.

Technologies include: devcontainers, nix, Docker-in-Docker, Kubernetes in containers, KubeVirt, Ceph, libvirt, KVM, Cilium, Zot, Dagger, Gitea, mise, Runme, Ghostty, tmux, Neovim, VS Code/Cursor, Talos Kubernetes, and more. It's a complete technologist's toolbelt in action.

You'll feel at home whether you're a terminal junkie, TUI connoisseur, and CLI ninja, or if you prefer a refined full IDE and click-to-run exercises. Join for hands-on technicals that break down complex bits into familiar patterns connected to the tangible hardware in your hands. Bring your laptop to follow along, or snap the QR code and try it later with friends. Take home the as-seen-on-TV toolkit and everything needed to hack your own cloud projects.

This session walks through how to build and train your own Small Language Model using open source tools. You'll learn how to select datasets, reuse BERT's tokenizer, design a smaller student model, and apply knowledge distillation to preserve accuracy. We'll share code, show common pitfalls, and help you deploy an efficient, real-world SLM that runs fast on modest hardware.

This session highlights our journey enabling all 3 new trends altogether, Flutter for AI Automotive experiences on RISC-V hardware. We explore the open-source components, performance optimizations, and system design needed to support responsive coming HMIs, and discuss how the community can contribute to building a fully open RISC-V AI automotive software ecosystem with Flutter popularity.

Building on the fundamentals of multi-GPU architectures, this talk dives deep into solving the 2D Burger's equation—a cornerstone of fluid dynamics. I will explore the critical relationship between grid resolution and time-to-solution, demonstrating how to effectively leverage multi-GPU compute to optimize performance for complex engineering and physics problems.

Imagine planning a trip where AI agents for flights, hotels, rental cars, and events collaborate seamlessly. Instead of repeating details across separate systems, these agents share information in real time—adjusting dates for a music festival or suggesting itinerary changes before booking. This talk explores how to create such shared multi-agent conversations using Asterisk, leveraging ConfBridge for audio mixing, AudioSocket for AI integration, and the Open Floor Protocol (OFP) for interoperability. By aligning with OFP, we enable AI agents to collaborate securely, share context, and manage interruptions, transforming digital interactions from transactional to collaborative.

Agentic workflows are rapidly becoming the next major shift in software engineering: systems that can plan, act, verify, learn, and repeat—turning “prompting” into production-grade automation. In this talk, I’ll share a practical framework for how developers can adopt agentic workflows responsibly, and why they unlock a new level of leverage for building and operating complex systems.   The goal is simple: help the Linux community build software systems that get better over time—secure by default, and fast by design today.

LLMs enable new perspectives in the analysis of narrative text from patient level hospital data. I will show how to carefully integrate them in medically sound pipelines to perform tasks as identifying disease activity, comorbidities, and risk factors, which were traditionally done manually and can now be scaled by orders of magnitude. I will highlight the usual cornerstone steps in these pipelines, distinguish reasoning-heavy from NLP-heavy applications, and showcase the key differences to consider in public pre-trained models as Llama and GPT-OSS and open-source software as Ollama and vLLM.

Over the Summer, I worked as an intern for Costco Logistics, where I spent the majority of my time developing an AI-driven project for their website and services. This experience gave me a firsthand insight into how major corporations are integrating artificial intelligence into their daily operations, and it shaped my understanding of the growing role that AI plays in modern logistics. In my presentation, I'll be sharing what I learned throughout this process, from technical skills that I gained to the broader trends that I observed. In addition to detailing the process of the project and its results, I will discuss how it and other novel AI endeavors transform the shipping industry around the world.

The session explores how personal health telemetry and open modeling can turn oral health from an afterthought into a lever for whole‑body wellbeing. Developed by 21D with research support from the Kwaai AI Lab, this open-source project uses Markov health‑state models and NICE‑aligned economic data to simulate how changes in periodontal status ripple through diabetes, cardiovascular risk, and long‑term healthcare costs.
 

If you’re a practitioner, developer, or data person, you’re probably being told to “just use AI”. But what does that actually mean?

What is “AI” really? Is it just ChatGPT, Claude, or Canva drawing pictures for you? 

This talk maps out the main building blocks of modern AI. I’ll discuss models and model architecture, training, embeddings, tokens, parameters, agents, RAG, hallucinations, “agentic” systems, and a few other terms you keep hearing.

Importantly, we’ll look at how these pieces are actually put together in real systems: where they’re genuinely useful, where they’re fragile, and where they’re oversold.

You’ll come out of the talk with a solid mental picture of how these ideas connect, what’s behind the buzzwords, and enough understanding to engage in a more informed way on your won AI journey.

A subset of signal leak attacks focusing on keyboards. This talk is compiled with open sources, no classified material will be discussed.

Attendees will learn about QEMU user mode, how it can be used to run applications compiled for other architectures as if they were natively built for the user's platform, how to configure their systems to make executables from all architectures executable by default, and how applications can be packaged for multiple applications using QEMU.

MySQL 8.0 was released GA in April 2018 so it turns 8 years old this year. In April 2026 it will receive its last patch update and then it will go end-of-life. I will run you through the deprecations that are gone in the next LTS release 8.4 and how too avoid breaking your existing tooling.

Kubernetes can be complex to troubleshoot, especially when it comes to remembering precise CLI commands or piecing together multiple tools. In this talk, we introduce the open source project Inspektor Gadget -  a CNCF sandbox project whose goal is to simplify Kubernetes observability. Leveraging the power of eBPF, Inspektor Gadget enables users to gain low level visibility into their Linux systems within the context they're familiar with like pods, nodes, and namespaces. In this presentation we'll show you the easiest way to use Inspektor Gadget for troubleshooting - the Inspektor Gadget MCP Server. This is an open source project designed to help users investigate and resolve some of the trickiest Kubernetes issues (including DNS!) by using natural language queries. With just a few thoughtful prompts, we can go from the chaos and frustration that comes along with troubleshooting issues to the sense of achievement that is derived from successfully identifying the root cause of an issue. Join us! 

Joshua Colp will review the project's current status, key developments, and community contributions, including recent advancements, challenges, and future directions of the Asterisk open source telephony framework.

AstriCon opens with an engaging panel discussion featuring key figures from Sangoma, the driving force behind Asterisk and FreePBX. This session will cover the significant achievements and challenges of the past year, industry trends, and exciting developments on the horizon.

The Atmosphere — built on AT — presents a new approach to social graphs. When you write data using Atmosphere APIs, such as by posting to Bluesky, that data is associated with your personal data repository. These personal data repositories can be hosted or migrated anywhere across the Atmosphere. Each Atmosphere app declares its own schema (Lexicon), and reads and writes its own set of fields. These fields can be read by any other app built on the Atmosphere, allowing users to both a) own and b) span their graphs across the network.

This talk will provide a demonstration of some fundamental AT technologies, including:

- "Sipping the Firehose" - working with the stream, a demo of creating records and have them pop right out
- “Getting backlinks with Constellation” - querying social interactions in real time, and building that data into different interfaces
- “Lexicon Authoring” - a discussion of best practices for creating additional schemas, with examples from other apps in the Atmosphere

Kubernetes is exceptionally good at keeping things running—and sometimes a little too good at hiding when they’re not. Pods restart, nodes reschedule, alerts auto-resolve, and dashboards stay green long after the system has started misbehaving.

In this 10-minute talk, we’ll walk through examples of Kubernetes failure patterns where missing observability signals and incomplete instrumentation allow small problems to quietly pile up until they become a production outage. We’ll show how infrastructure-heavy dashboards—and “AIOps” systems confidently declaring “this is fine”—create blind spots that delay human intervention.

Using OpenTelemetry as a practical example, we’ll dig into how to instrument for behavior instead of resources, correlate metrics, logs, and traces by default, and surface early warning signs before users feel the impact.

Because barns don’t burn down because of one loose nail—but modern production systems absolutely can.

Over the last three years, Bazzite has established itself as one of the best entry points for Windows gamers in their journey to Linux. An OS that is foolproof, ready to game out of the box, and features the latest drivers built-in. It provides a premier gaming experience for all PC form factors: Desktop, Laptop, and Handheld. Bazzite is redefining the potential of the Linux desktop by harnessing the full power of Cloud Native patterns. By embracing an image based approach, Bazzite transforms the operating system from a complex assembly of packages into a seamless, versioned, and universally deployable container image that is exceedingly difficult for an end user to break. 

In this session, we will show you how to incorporate our model into your projects and demonstrate how the model is fundamentally setting the standard for the future of personal computing. We'll also chat about how you can help the entire Linux gaming ecosystem by contributing to the Open Gaming Collective (OGC).

Static analysis tools are fast, scalable, and widely used in modern software workflows, but they struggle to reason about runtime behaviors in complex embedded systems. This talk focuses on how symbolic execution can be used as a complementary technique to explore deeper execution paths and uncover subtle bugs that traditional static analysis often misses. We will explain the core concepts, key challenges like path explosion, practical mitigation strategies, and real-world case studies involving embedded Linux applications.

AI coding assistants have made engineers 10x faster, but when entire teams use them, a new problem emerges: architectural chaos. Each developer gets working code from their AI, but without shared architectural context, you end up with conflicting patterns, duplicate services, and unmaintainable systems. This talk introduces SpecMind, an open-source tool that enables spec-driven vibe coding by automatically analyzing your codebase, generating architecture documentation, and keeping design decisions aligned across your team. Learn how to maintain architectural consistency while moving fast with AI, and see a live demo of the analyze, design, and implement workflow that keeps teams building together instead of in isolation.

iptables and nftables are the standard for Linux packet filtering: well-documented and widely understood. But at high bandwidth, BPF is the fast alternative, although it means writing C and fighting the BPF verifier.

bpfilter bridges the gap: an iptables-like DSL that compiles to BPF bytecode.

This hands-on talk covers Linux filtering tradeoffs, introduces bpfilter, and demos real-world usage from writing rules to debugging.

bpftrace provides a quick and easy way for people to write observability-based eBPF programs, especially those unfamiliar with the complexities of eBPF. We always claimed bpftrace was a "high-level tracing language" for Linux even though it was missing the basic features of a language: composability, primitives to avoid code duplication, and even proper loops. It was also trailing behind the larger, upstream BPF feature set. This talk is about how we're working to transform bpftrace from a box of tools (one-liners) to a language for making new tools.

https://bpftrace.org/

Building a powerful open source brand isn’t about marketing hype. It’s about principles. This session unpacks how transparent values, genuine community engagement, and a user-first mindset create lasting trust and loyalty. Drawing on DBeaver’s evolution from a 2010 hobby project to a tool with over 10 million users, discover how prioritizing help over hype, collaborating openly with users, and fostering a likable, responsible voice transforms contributors into ambassadors and positions open source projects for sustainable commercial success... without sacrificing community goodwill.