DevConf.US 2025
as the organisers put it
DevConf are free and open, annual, open-source community conferences for community contributors, sponsored by Red Hat.
who it's for
platform engineers, cloud native engineers, machine learning engineers, site reliability engineers, devops engineers, developer tools engineers, security engineers
topics
Kubernetes native LLM inferencing; continuous batching for inference; KV cache page retention; GPU direct RDMA transfers; OpenShift installer evolution; eBPF observability and telemetry; WebAssembly for cloud and edge; Triton kernel caching with OCI; RAG deployment with Llamastack; Kubernetes cost optimization strategies; Ansible run capture with ARA; podman capabilities and retrospectives
| title | author | length |
|---|---|---|
| llm-d: Kubernetes Native Distributed Inferencing | Robert Shaw | 33m |
llm-d serves LLMs by marrying VLM-style inference with Kubernetes: it implements continuous batching to pack per-step token work, manages massive KV cache page… | ||
| Unifying Developer Experience Across Virtual and Cloud Native Worlds | Heber Romero Tellez, Francisco Zompa, Jorge Varela, Francisco Javier Monroy Garcia | 33m |
Argument: unify VM and container workflows so developers stop context switching and security blind spots. Shows a working stack using Crossplane to provision… | ||
| ARA Ansible for the team | Kirill Satarin | 14m |
ARA Records Ansible is a lightweight web service plus a Python callback plugin that captures every playbook run, saving playbook files, host facts and task… | ||
| WebAssembly's Journey Beyond the Browser | Matthew Flug | 19m |
WebAssembly runs on Wasmtime and WasmEdge in cloud and edge runtimes, uses the WASI standardized 2023 API set to safely call networks and files, and integrates… | ||
| Transmitting longwave radio signals from wired headphones | Dominic Delabruere | 14m |
Replicates regional longwave time broadcasts using a laptop or phone and wired headphones by driving the headphone driver at 20 kHz (or other audio… | ||
| GPU Programming with Triton Kernels | Kyle Yu | 08m |
Practical primer on writing Triton GPU kernels with concrete results: shows how kernels use tl.load/tl.store and block-offset indexing for vector add, how a… | ||
| An open cloud for european researchers, staff and students | Colm Dunphy | 15m |
An experimental European research cloud repurposes a liquid cooled supercomputer into three OpenShift multi node clusters running RHEL on bare metal and… | ||
| Reusable UI Components in a Framework Multiverse | Hadar Geva | 14m |
Shows how existing React components become native web components and run unchanged in plain HTML or Vue by wrapping them with a small bridge. Builds a video… | ||
| Portfolio Marketing Metrics and Insights | Hannah Choe | 07m |
Built a dashboard that surfaces engagement and performance for Sales Plays content from the Red Hat Content Center, using SQL and pivot tables to explore raw… | ||
| Zero Trust Visualization | Nathan Strahs | 07m |
A NIST security profile runs on an OpenShift cluster producing XML that the tool converts into OSCAL; the visualizer ingests OpenShift compliance operator… | ||
| Perf Characteristics of Userspace & Kernel-Space Datapaths in Container Networking | Yanhui Ma | 14m |
Proof-of-concept VDUSE OVS-DPDK userspace datapath is implemented for OpenShift using OVS-DPDK, a VDUSE kernel driver (control via a character device) and… | ||
| Probabilistic Analysis of WCETs on RT Systems: When Your Worst Case Isn't Worst | Aseef Imran | 08m |
A measurement toolkit applies EVT and the generalized Pareto distribution to PREEMPT_RT latency traces, selects thresholds via a stability plot, and reports… | ||
| Born Agile: Why Engineers Resist What Comes Naturally | Allison King | 24m |
Argues engineers are born agile, tracing specific parallels: minimal MVPs (baby crawling → first race), iterative demos, and a real demo where a child earns a… | ||
| Unveiling Remote Code Execution: How Vulnerabilities Lead to System Takeovers | Suyash Nalawade | 28m |
Walks through three real CVEs with code-level detail and a demo showing exploit chains and fixes: CVE-2024-47076 (cups-filters) lets a malicious print server… | ||
| Breaking GitOps Bottlenecks: The Power of Sharding | Anuj Tyagi | 17m |
Single-controller GitOps hits resource limits as teams and changes scale, and simply bumping CPU/memory is only a stopgap; sharding controllers instead splits… | ||
| Impact Beyond Code: A Guide to Technical Leadership | Namita Soman, Preethi Thomas | 29m |
Teams at Red Hat run container tooling like podman, dnf and rpm and the claim here is that AI agents as superpowers let a single IC do work that once needed a… | ||
| Market Requirements Document | Cherilyn Kim | 08m |
A 19-page MRD documents market needs in North American healthcare IT, identifying a surprising 19-day prior authorization delay for treatments, recommending… | ||
| Introduction to Perses, an Open Source Observability Dashboard | Jenny Zhu | 15m |
Perses is an open-source dashboard built for GitOps-native workflows and accepted into the CNCF sandbox in August 2024. It visualizes telemetry from… | ||
| Simplifying Cloud Native Application Development with Kind | Alyssa Goins | 08m |
Kind runs local clusters by using containers as nodes, Kubernetes in Docker, and Kubeflow Pipelines uses Kind for CI and local development. Workflow is: run… | ||
| 8Knot Chatbot | Alvin Yu | 08m |
A chatbot for 8Knot that recommends which of the 30–40 visualizations to show and then uses LLM function-calling to populate Plotly graph parameters; graph… | ||
| Five podman features you didn’t know about or that you should consider start using | David Darrah | 15m |
Desktop installs in a minute on macOS/Windows using a small VM (libkrun/WSL) and exposes a GUI with image/pod management and plugins including a Podman Desktop… | ||
| Podman Health Check: A Data-Driven Look at Community, Performance, and Trends | Mohan Boddu, Ashley Cui | 31m |
Podman reports concrete metrics: 3,300 Matrix members and 3,200 Discord (~6,000 distinct), 15% Discord retention, 25 authors and 111 commits to main last… | ||
| xDSnap: A Lightweight Tool to Debug Envoy and Consul Dataplanes with Confidence | Mark Campbell-Vincent | 11m |
xDSnap is a Go-based open-source tool that captures Envoy config dumps, stats, pod logs and optional TCP dump pcap from Consul dataplanes and packages them as… | ||
| Streamline Linux & Windows Nodes debugging in K8s with new Node Log Query feature | JR Valdes | 26m |
Node Log Query (KEP-2258, in Kubernetes 1.30 beta with GA work ongoing) lets clusters retrieve files and Windows Event Log entries and query named services… | ||
| Things I Wish I Knew When I Became a Manager | Micah Abbott | 29m |
Covers the first 90 days up to six months: expect constant approvals in broken HR systems, uncomfortable money conversations about bonuses and raises, and… | ||
| UV Migration | Moryan Tchoumi Noukoua | 09m |
Migrates many notebook images from Pipenv to UV by converting Pipenv locks into pyproject.toml, changing Dockerfiles to pip-install UV, and adding GitHub… | ||
| AI in the Terminal: Empowering Developers with CLI-Based AI Apps | Aditya Gidh | 15m |
A small CLI app proves AI in the terminal can be practical: a three‑command tool (a log analyzer, web performance auditor, and web accessibility auditor)… | ||
| ZenZone: AI-Powered Peace of Mind | KEERTHI UDAYAKUMAR | 15m |
A prototype web app uses text, voice and facial cues to flag early depression, integrating an LLM comfort bot (Llama 2 tested), facial emotion features from… | ||
| Why post-quantum cryptography? | Dominic Delabruere | 14m |
Public-key crypto becomes breakable when quantum machines can factor large integers and solve discrete logarithms efficiently; experimental milestones: IBM… | ||
| Bootstrapping Application & Op Development with Devcontainers & Grafana LGTM Stack | Tyler Auerbeck | 35m |
Shows a reproducible local dev loop that boots an app plus observability stack using Devcontainers + Docker Compose, Grafana (Loki, Tempo, Prometheus/Mimir)… | ||
| No UI, Still a Journey: Mapping UX for the Unseen | Tiffany Nolan | 16m |
Software systems fail because back-end naming, APIs and logic ignore the user's path, not because the UI looks bad. The talk claims mapping muddled map… | ||
| Auto-tuning vllm | Rehan Samaratunga | 10m |
Autotuning vLLM experiments use Optuna (TPE, BoTorch, grid) and MLPerf benchmarks to search config space for Qwen3-32B-FP8 and Qwen3-30B-A3B-FP8, comparing… | ||
| Post-Quantum Cryptography for Digital Signing | Arthur Savage | 36m |
Post-quantum signing moves from theory into tooling: NIST-selected algorithms like ML-DSA and competing choices such as SLH-DSA are being integrated into TLS… | ||
| Slide Karaoke! | Jeff Ligon | 29m |
A live slide‑karaoke session runs from a Python script that auto‑advances random slides 15 seconds per slide, volunteers improvise humorous narrations about… | ||
| Building easily deployable RAG applications with Llamastack | Peter Zhao | 11m |
A working GitHub RAG assistant is built: repos are cloned, documents chunked, embedded and stored in Postgres + pgvector, and semantic search returns source… | ||
| P2CODE: A programming platform for developing and deploying edge and IoT apps | Clodagh Walsh, Paul Power | 22m |
P2CODE delivers a Kubernetes operator scheduler built on Open Cluster Manager that accepts a custom P2CODE scheduling manifest, auto-bundles related manifests… | ||
| Accelerating Boot Time: Lessons from a New Platform | Ed Chong | 21m |
Measures wall-clock boot with the ARM CNTVCT counter and serial capture (Grab Serial), runs 50+ samples, and adds early-kernel patches when needed to anchor… | ||
| Finding your place in an industry that seems built for someone else | mb Pierce | 35m |
Tech workforce numbers shock: 27% women, 4% Black, 8% Hispanic, neurodivergent folks ~20% of population but ~3% in tech, disability ~6%, and over half of many… | ||
| Getting Started in Open Source | Amy Marrich, Shaun McCance | 32m |
Contributions on GitHub/GitLab or self-hosted repos (OpenStack on openinfra.dev) start with small, tagged issues like good first issue or QR-linked Podman… | ||
| Moving from Management to Individual Contributor, a journey | Jeff Ligon | 39m |
Red Hat builds a safety-focused Linux for cars that must make a braking decision in 500 microseconds and compete with QNX in 98% of cars. The project hires… | ||
| Cut Costs, Not Performance: Implement Kubernetes Cost Optimization Strategies | Ankur Singh | 14m |
Three concrete tactics reduce Kubernetes cloud spend with measurable results: cluster resizing (remove nodes when overall CPU and memory sit ~25–45%), instance… | ||
| How I made a powerful cache system using Go | Sylvain Combraque | 37m |
Built a Go HTTP cache called Souin that plugs into Traefik and Caddy, backed by multiple storages (badger, filesystem, Redis and others) and deployed via… | ||
| Learn How to Run an LLM Inference Performance Benchmark on NVIDIA GPUs | Ashish Kamra, David Gray, Samuel Monson | 33m |
Step‑by‑step recipe shows a repeatable Fedora workflow: enable RPM Fusion, install akmod-nvidia, verify with nvidia-smi, add Podman 5.x and NVIDIA Container… | ||
| A Scrappy UXer's Tale of Effective Low-Tech, Low-Prep, Quick Turnaround UX Research | Mary Shakshober | 33m |
Claims a repeatable, low‑prep method that uncovers user mental models about GUI terms using a bare word slide‑deck template, two 1:1 sessions plus a… | ||
| Smarter RAG, Smaller Bill: Optimize for Performance and Price | KEERTHI UDAYAKUMAR | 14m |
RAGCache caches LLM key‑value tensors in GPU and host memory using a dynamic knowledge tree and a prefix-aware dual-size-frequency replacement policy (PGDSF)… | ||
| An All Code Introduction to Domain Driven Design and Hexagonal Architecture | Jeremy Davis | 27m |
Live code builds a Quarkus REST app running on Podman that replaces CRUD with DDD plus Hexagonal wiring: a Thought domain object enforces business rules and… | ||
| Greener Intelligence: Reducing AI’s Carbon Footprint | Twinkll Sisodia | 14m |
Massive model training produces concrete emissions: training a large language model equals the lifetime CO2 of thousands of cars, and serving queries totals… | ||
| Putting AI Agents to Work: A New Era of Open Source Connectivity with MCP | Hema Veeradhi, Yu An | 36m |
Claim: MCP makes agents modular and interoperable by standardizing tool metadata, auth, and calls. The team builds a cloud-native stack with vLLM, Llama Stack… | ||
| OpenShift Installer: Past, Present, & a Potential Future | Patrick Dillon | 34m |
openshift-install is a directed acyclic graph that turns a minimal install config into Kubernetes manifests and giant ignition files, and it provisions cloud… | ||
| An AI/ML-Driven Approach to Platform Lifecycle Management | Prithvi Raj Venkat Raj, Justin Winter | 35m |
A Kubernetes-native control plane called Cordant manages metal-to-model stacks and multicluster multicloud ops, using ClusterAPI + k0s for templated cluster… | ||
| eBPF: Revolutionizing Observability and Telemetry for Modern Applications | Mike Dame, Eric Horsman | 37m |
eBPF runs as a verified virtual machine inside the Linux kernel and is used here to attach kprobes and uprobes to live processes, compile small C/Rust… | ||
| Scanning for FIPS : Building confidence in Compliance from the Container up | Yashvardhan Nanavati, Nate Marsella | 31m |
Pipeline adds a post-build step that unpacks images with umoci and runs check-payload (scan-image/scan-local) to statically inspect ELF binaries, enforce… | ||
| Is Frontend Still a Thing? A UI Reset in the Age of AI Disruption | Keren Fanan | 33m |
A photo-platform prototype—created with Lovable and turned into functioning code by Gemini-generated HTML/CSS—is copy/pasted into an open-source integration… | ||
| Upstream Communities - Key Partners in UX Research | Vince Conzola, Jenn Giardino | 30m |
Red Hat UX engages the upstream Kubeflow community to validate a proposal: give users a model registry latest-version shortcut. Recruitment uses a screener… | ||
| Image mode for AI inference across the hybrid cloud | Mayur Shetty, David Duncan | 32m |
Demonstration builds a RHEL bootable image by containerizing a Ramal Llama model and chatbot with Podman, then using the bootsy image builder to produce a raw… | ||
| DevToDeploy: Integrate Your PM & CICD Tool For Systematic & Faster SW Delivery | Ankur Singh, Shaheen Sayyed | 35m |
Claims a production‑proven pipeline that removes manual release steps by wiring Jira webhooks to CI/CD: GitLab CI triggers builds and tests, updates Jira… | ||
| We can automate that! | Zack Zlotnik | 17m |
Shows a concrete workaround: an internal ticket feed without an API is reconciled to Google Calendar by a Python scraper that parses predictable messages with… | ||
| Zero to GitOps: Automating RAG Workloads with Backstage Templates | Rose Crisp | 37m |
A single template automates GitOps repo scaffolding, Argo CD manifests, and Helm chart deployments on Red Hat Developer Hub (Backstage) running on OpenShift… | ||
| 7 Years of Podman: A Retrospective | Matthew Heon | 36m |
Project starts in 2015, adds rootless within months, then ships weekly to reach basic parity with Docker; Podman 2.0 implements a REST-compatible Docker API… | ||
| Sculpting Subspaces: Constrained Full Fine-Tuning for Continual Learning in LLMs | Nikhil Shivakumar Nayak | 29m |
A full-parameter continual-learning method leverages adaptive SVD to split each weight matrix into a frozen high-rank subspace and a tunable low-rank subspace… | ||
| SDG_Hub: An Open-Source Toolkit for Synthetic Data Generation & LLM Customization | Shivchander Sudalairaj | 36m |
SDG Hub is an open-source toolkit that composes prompts, blocks, and flows into reusable, YAML-or-Python pipelines that scale automatically and wrap LLM calls… | ||
| Characterizing AI inferencing stacks | Ryan Malone | 10m |
Project simulates stacks by running Granite Vision 3.2 (2B) on an Nvidia Ampere GPU in a Jetson AGX Orin with real 9.4, swapping only the runtime to compare… | ||
| Disease Detection in the Age of Quantum Computing | Andrew Miller | 34m |
Claims a working pipeline that trains a hybrid classical–quantum classifier to detect sepsis, encoding patient features with a ZZ feature map and training a… | ||
| K8s Under Fire: Igniting Performance & Burning Through Bottlenecks with Kube-Burner | Sai Sindhur Malleni, Vishnu Vardhan Reddy Challa | 30m |
Kube-burner is a CNCF Sandbox tool that orchestrates large-scale K8s workloads, creates namespaces and thousands of objects (demo runs 10 iterations creating… | ||
| What's new with RamaLama | Daniel Walsh | 37m |
Project demonstrates running models locally in containers with Podman, auto-detecting GPU type and pulling a matching image, dropping privileges for isolation… | ||
| AI Agent Orchestration for Enterprise Arch with AWS Bedrock, Flyte & LangGraph | Toufic Boubez | 31m |
Built an enterprise AI orchestration stack that combines AWS Bedrock (Claude 3) for inference, LangGraph for multi‑agent state and execution plans and Flyte… | ||
| Beyond the Buzz: Discovering High-Impact Use Cases in the Generative AI Era | Aditi Saluja | 20m |
Generative AI pilots 95% of GenAI pilots die because teams favor flashy demos and thin wrappers over ROI and defensible models; common failures include… | ||
| Agentic Reasonings with Reinforcement Learning | Andy Xie | 15m |
Applies reinforcement fine-tuning to get LLMs to use tools and past state instead of shallow pattern matching, building a Wordle agent that learns to prefer… | ||
| Training AI Models on Massive Datasets | Julio Faracco | 31m |
Framework DASF combines Dask lazy loading, RAPIDS GPU acceleration and direct-GPU IO (Kvikio / GPU Direct Storage) to train models on seismic SEGY data… | ||
| LLMs on the Edge: The Future of On-Device Intelligence | Rakesh Musalay | 23m |
Shows a container-native, on-device pipeline—Vision LM → RAG → LLM—running rootless with Podman and GPU CDI on Jetson Orin Nano (8 GB GPU) under RHEL Edge… | ||
| Obituary - Lessons learned with a career in software? | Daniel Walsh | 47m |
Career summary spans 24.5 years at Red Hat, starting with an NSA-funded $5M push to upstream SELinux and inventing the SELinux sandbox and MCS-based VM… | ||
| Extending Quality Management with Audio, Video, and KVM using Podman Quadlets | Yariv Rachmani, Douglas Schilling Landgraf | 35m |
Shows how Quality Management zones gain host devices and virtualization using Podman Quadlets and dynamic OCI hooks to attach GPUs, audio, Wayland compositors… | ||
| Acing Distributed Container Management | Prithvi Raj Venkat Raj, Justin Winter | 35m |
An AIOps-driven platform called Cordant acts as a mothership cluster built on ClusterAPI with templated YAMLs to avoid repetitive cluster specs, Sveltos for… | ||
| Intelligent Pipelines: Bringing AI/ML into DevSecOps Automation | Heber Romero Tellez, Francisco Zompa, Jorge Varela, Bernardo Guerrero, Francisco Javier Monroy Garcia, Aldo Luevano | 30m |
Claims an MCP/NCP platform that embeds ML and agents into CI/CD to run predictive vulnerability detection and smart incident response, implemented as a… | ||
| CoreOS Demystified – The Specialized OS Powering OpenShift and Containers | Renata Ravanelli | 24m |
Explains that Fedora CoreOS (FCOS) and Red Hat CoreOS (RHCOS) are minimal, immutable OSes for containers with automatic two-week updates and a single-image… | ||
| Cloud-Native Model Serving: vLLM's Lifecycle in Kubernetes | Cedric Clyburn, Roberto Carratala | 36m |
vLLM runs as a Kubernetes-native inference server that maximizes throughput by combining quantized models, KV cache/prefix caching, and dynamic batching; a… | ||
| Which Is To Be Master? Understanding LLM Tokenization | Erik Erlandson | 27m |
Compares concrete tokenizers—Byte Pair Encoding, WordPiece, and unigram tokenizer—and trains tokenizers from corpora, then runs HMM surrogates to measure… | ||
| Don't tell me RAG is easy | Major Hayden | 36m |
A team builds a RAG-backed product and finds it’s a process, not a checkbox: they use keyword + vector hybrid search, embeddings and tools like DocLink and… | ||
| From Cold Start to Warp Speed: Triton Kernel Caching with OCI Container images | Maryam Tahhan, Alessandro Sangiorgi | 25m |
Packaging Triton-generated LLVM/CUBIN kernels as signed OCI images and pre-extracting them to nodes reduces JIT time and halves model cold-starts. A toolset… | ||
| Unlocking Business Insights with Natural Language and Generative AI | Vijaya Bashyam, Neerju Gupta | 35m |
IBM’s Semantic Layer ingests catalog metadata once from millions of legacy tables and 100,000+ columns, runs metadata enrichment with Granite 8B and vectorizes… | ||
| Generative AI Model Data Pre-Training on Kubernetes: A Use Case Study | Santosh Borse | 34m |
Shows a production pipeline using Kubeflow Pipelines + KubeRay to spin up Ray clusters, run deduplication, tokenization and content classification, then tear… | ||
| Demystifying Confidential Computing: An update for enthusiasts & professionals | Jake Correnti | 23m |
Covers concrete stack and status: guest support lands in Linux 5.19, SEV-SNP host support in 6.11 and TDX host in 6.16; QEMU supports SEV in 9.1 and TDX in… | ||
| The 10 steps of Container Security: It's not that hard ! | Boaz Michaely | 35m |
Concrete checklist: ten essential risk areas are enumerated and mapped to controls using open source Stackrox / Red Hat ACS, with image scanning in CI, signing… | ||
| Getting real DevOps ROI with your Generative AI | Allen Gilbert Andrews | 41m |
Shows a deployed GenAI ops stack that runs on OpenShift/k8s and is already live in five states handling 1.5M transactions/day via a product called Health… | ||
| Test-Driven Development: It's easier than you think! | Eric Deandrea | 30m |
A demo builds a small Java 21 Quarkus app (REST + JPA + Hibernate + PostgreSQL) using Quarkus Dev Services to auto-start a Postgres container and Continuous… | ||
| Keys, Certs and Vault: From DevSecOps to Post-Quantum | Chaitanya Challa | 29m |
Expired certificates cause real outages; a recent incident cost an evening of root‑cause work and would have been prevented by automated renewal and… | ||
| Why and How to Choose Arm64 for your Cloud Applications | Dave Neary | 36m |
Arm64 instances (Ampere) claim 20–30% lower core-hour cost and reports show 30–47% better performance‑per‑price for real workloads; case studies include object… | ||
| How I trained an AI Model to Beat the 1990's Arcade Game Double Dragon | Christopher J Nuland | 37m |
A distributed training pipeline uses OpenShift + KubeRay to run ~100 emulator instances across 24 A100s, training a PPO-based agent that reliably clears level… | ||
| Unlocking Smarter AI with Inference-Time Scaling | Guangxuan Xu, Kai Xu | 28m |
Claims that running extra compute at test time boosts accuracy without retraining and that a Python library makes those methods plug-and-play. Supports… | ||
| Docling: Get your documents ready for gen AI | Michele Dolfi, Peter Willem Jan Staar | 35m |
Docling is an open source Python library that converts PDFs, DOCX, PPTX, HTML, images and Markdown into a lossless DoclingDocument and easily exportable… | ||
| Language Model Post-Training in 2025: an Overview of Customization Options Today | Mustafa Eyceoz | 36m |
Overview of concrete options and what actually works: standard SFT and parameter‑efficient methods like LoRA / QLoRA / OLoRA let you add small delta weights or… | ||
| nothing matches | ||