Andy Anderson
IBM Research, KubeStellar Community Maintainer
Stamford, Connecticut, United States
Actions
Andy is an experienced software architect with a strong track record of leading research and engineering teams on hybrid cloud and Kubernetes-based platform development. He has a wealth of experience in software development, cloud computing, and data analysis.
Andy is particularly adept at combining continuous integration and delivery practices, containerization and orchestration technologies, and cloud computing products and services for use with platform engineering efforts.
Area of Expertise
Topics
Your Codebase Is the AI Model: A Maturity Framework for Autonomous Development
AI coding tools are everywhere but most teams plateau at "prompt and review." This talk presents the AI Codebase Maturity Model, a 6-level framework defining maturity not by what the AI can do but by what your codebase encodes — instructions, tests, metrics, feedback loops. Each level depends on the previous; you cannot skip. We walk from Level 1 (ad-hoc prompting) to Level 6 (autonomous multi-agent orchestration), validated against a CNCF project. The central insight: automating a repo with AI agents demands the same rigor as running a large open-source project — testing, coverage gates, governance. A solo dev with AI agents IS a large team. Takeaways: (1) flaky tests in autonomous workflows are dangerous, not annoying, (2) intelligence lives in the infrastructure, not the model, (3) full autonomy needs more human judgment, not less. We demo Hive, the open-source Level 6 implementation. Attendees leave with a self-assessment and artifacts to reach each level.
Single Cluster: Easy Peasy. Multi-Cluster: Real Pain? We feel you
It starts with a single Kubernetes cluster - clean, contained, and calm.
Then comes growth. Teams expand, workloads spread, and soon clusters bloom across edges, clouds, and on-prem worlds. What once felt like orchestration begins to sound like chaos: drifting configs, tangled policies, and reliability that rolls like dice. Innovation slows under the weight of its own sprawl.
This panel brings together a diverse mix of voices - CNCF Ambassadors, independent practitioners, and freelancers from around the globe. Together, they unpack the real pains and promise of scaling beyond one cluster.
We’ll explore open-source paths that keep freedom at the core:
1. KubeStellar and Karmada for multi-cluster federation
2. Cluster API for provisioning at scale
3. Istio for service mesh consistency
Scaling AI Inference Across Remote Clusters with VLLM
AI models are growing in size and complexity, making efficient workload distribution essential for scalable inference. This session explores how to deploy and manage AI inference workloads across remote Kubernetes clusters using VLLM, an optimized solution for high-throughput inference.
Attendees will learn:
- How VLLM improves AI inference performance
- Techniques for distributing workloads dynamically across remote clusters
- Strategies for balancing latency, compute availability, and cost
This talk is ideal for AI engineers, platform architects, and open-source enthusiasts looking to optimize AI inference across distributed infrastructure.
Overcoming Multi-Cloud GitOps Hurdles: Building a Unified Multi-Cloud AI/ML Platform in Kubernetes
In this session, we explore the GitOps challenges involved in building a unified platform for multi-cloud Artificial Intelligence (AI) and Machine Learning (ML) workflows on Kubernetes. Discover the intricacies of integrating key open source projects—Open Data Hub, Codeflare SDK, KubeStellar, KubeFlow Pipelines (KFP), and Multi-Cluster Application Dispatcher (MCAD). Learn about our source control challenges, such as handling secrets, automation hurdles with service accounts, and post-deployment configuration to achieve a 12-minute deployment cycle. Gain insights into the challenges of orchestrating deployment tasks across clusters and clouds, aiming to engineer a consolidated and scalable multi-cloud Kubernetes ecosystem for AI/ML operations.
From Classroom to Cloud: Bridging Academia and the Open Source Ecosystem
This session introduces **Interns for Open Source (IFOS)**, a program that offers undergraduate and graduate Computer and Information Sciences students hands-on experience with open source projects for academic credit. Over 10 weeks, students bridge classroom learning and real-world application by contributing through issue tracking and pull requests. Their fresh perspectives provide open source communities with valuable feedback, usability insights, and rigorous testing. Students sharpen technical skills, learn professional workflows, and build portfolios. Open source projects benefit from innovative ideas and unbiased input. Attendees will learn about the program structure, its benefits for students and open source communities, and how to get involved.
AI Agents Everywhere: Hybrid and Edge Deployments with Kubernetes Multi-Cluster Management
As AI agents increasingly move from centralized cloud environments to the edge, managing their configuration and lifecycle across diverse, distributed infrastructures becomes a major challenge. This talk explores how CNCF projects support scalable multi-cluster configuration management for hybrid and edge AI workloads, enabling consistent and resilient operations across diverse environments. We’ll dive into patterns and best practices for deploying agents, models, and tools in connected and partially disconnected environments, addressing real-world constraints like intermittent connectivity and resource limitations. Attendees will leave with a robust architecture for hybrid edge AI, practical GitOps workflows for safe rollouts, and a clear decision framework for choosing the right multi-cluster tooling.
This talk is for engineers taking AI out of the datacenter and into the real world. Stop wrestling with bespoke scripts and learn to manage your agent swarm the cloud-native way.
From Idea to Release in 6 Weeks: Automating CI/CD for llm-d with Pipelines and Helm
In just six weeks, we took the llm-d open-source project from idea to release, navigating unexpected GitHub organization and repository name changes, as well as registry organization updates—while keeping CI/CD pipelines running without interruption. With 100 contributors restricted to non-cluster-admin permissions in Kubernetes, we implemented a namespace onboarding system and a GPU reservation system to ensure smooth operations. Using Kubernetes' Pipelines as Code and GitHub integration, we automated the entire process from code commit to AI model deployment, scaling across multiple clusters. In this talk, we’ll show how we adapted to these challenges, maintained pipeline continuity, and ensured proper access control across contributors. If you’re looking for strategies to manage non-cluster-admin users, automate complex workflows, and overcome unexpected disruptions in CI/CD, this session will provide actionable insights on how we pulled it all off in just six weeks.
Getting Started with Multi-Cluster Kubernetes: A Beginner's Guide to Fleet Management
Managing a fleet of Kubernetes clusters can be challenging, especially across diverse environments like edge locations, cloud, and on-premises. In this session, we’ll introduce the basics of multi-cluster management and explore how to simplify your workflows. You’ll learn how to deploy and manage Kubernetes across various distributions (e.g., kind, OpenShift, HyperShift) and keep edge environments in sync with your broader infrastructure.
We’ll discuss strategies for managing clusters without tightly coupling configurations, enabling flexibility and customization. By leveraging modular and abstracted approaches, Kubernetes operators can maintain an efficient setup that adapts to their needs, whether managing a few clusters or a global fleet. You’ll leave with practical insights into managing Kubernetes at scale, reducing complexity, and improving scalability and flexibility in your multi-cluster environments.
Giving an AI Assistant Root Access to Your Clusters: What Could Go Wrong?
AI assistants can now query your Kubernetes clusters, diagnose failures, and apply remediations — but should they? We built an MCP bridge that lets AI providers interact with live multi-cluster environments through a local agent, then watched what happened when operators used it. This talk covers: (1) connecting AI models to live cluster state — why a local agent with the user's own kubeconfig matters more than a service account, (2) guided missions that walk operators through installing CNCF projects across clusters, and why pre-flight checks and rollback are non-negotiable when AI suggests the steps, (3) the trust boundary — operators want AI help but not kubectl apply without review, and how we landed on "suggest, confirm, execute," (4) failure modes in production: AI recommending actions on the wrong cluster, hallucinating resource names, and the guardrails we added after each. Attendees leave with patterns for safely integrating AI assistants into Kubernetes operations.
Interns for Open Source (IFOS): Bridging Academia and Real-World Open Source Contributions
This session introduces Interns for Open Source (IFOS), a program designed to give undergrad and graduate students in Computer and Information Sciences practical, hands-on experience with open source projects for academic credit. Over a 10-week period, students apply theoretical knowledge to real-world challenges, contributing to open source projects through issue tracking and pull requests. This unbiased feedback from new users is invaluable, providing fresh perspectives and rigorous testing for open source projects. Attendees will learn about the program structure, benefits for students and open source communities, and how to get involved.
I’m afraid you can’t get there from here! WHAT IF … the Cloud can’t reach your Edges? Edge-Native
The CNCF IoT Edge Working Group recently published a new white paper, "Edge Native Application Design Behaviors," which will be explored in detail in this session. We'll discuss how to preserve the cloud experience - its ease of use, speed of deployment, familiar and versatile toolchains - when building and running applications on Edges that can't be treated as extensions of the cloud. These are edges that face connectivity or resource constraints, are "hidden" behind multiple layers of NAT or a non-transparent proxy, or are simply isolated for policy or security reasons.
By following edge native design principles that complement and extend cloud native, we can make even these remote, autonomous edges feel like the cloud. Let's explore these edge-native design principles together.
Multi-Cluster Configuration Management with KubeStellar
KubeStellar is a flexible solution for challenges associated with multi-cluster configuration management for edge, multi-cloud, and hybrid cloud
Mastering Canary and Blue-Green Deployments in Multi-Cluster Environments with AWS EKS
This session explores canary and blue-green deployment strategies in multi-cluster environments using AWS EKS, external-dns, Route 53, and KubeStellar. Drawing from the speaker's experiences, attendees will gain practical insights into managing clusters across different VPCs and regions.
Canary deployments gradually introduce new versions, ensuring stability, while blue-green deployments maintain two identical production environments for comprehensive testing. The presentation will highlight specific challenges faced during deployment on AWS EKS with ALB and offer solutions based on real-world experiences. Despite following AWS documentation, various hurdles were encountered due to gaps in EKS-specific knowledge, leading to unexpected results. This session will demonstrate how opensource projects like KubeStellar can facilitate successful canary and blue-green deployments, helping attendees avoid similar pitfalls.
Leveraging Kubeflow Pipelines, Argo Workflows, and KubeStellar for AI Data Pre-Processing at Scale
AI has become a vital component of many organizations' value propositions. To prevent unintended data exposure, businesses are opting to pre-process data for AI in-house. Many are successfully building singular pipelines for data pre-processing on laptops or single clusters. This presentation explores how Kubeflow Pipelines, combined with Argo Workflows and KubeStellar, offer a solution for organizations to achieve scale for their AI data preprocessing needs utilizing multiple Kubernetes clusters.
Attendees will gain insights into the importance of in-house data pre-processing for AI, the challenges organizations face, and the solution provided by Kubeflow Pipelines, Argo Workflows, and KubeStellar. The presentation will include practical implementation tips and real-world examples, enabling attendees to harness the power of these tools effectively.
Lessons from CNCF KubeStellar's Agentic Workflow Experiments
Open source practitioners from CNCF projects and GitHub share early findings from experiments with AI agents that execute GitHub tasks through natural language requests—transforming how maintainers manage open source workflows.
We'll demonstrate real agentic workflows in action:
- Automated contribution tracking and reporting across 19 maintainers
- PR review summaries and triage recommendations
- Issue classification and labeling at scale
- Documentation updates triggered by code changes
- Dependency management and security patch workflows
What seemed straightforward quickly exposed critical challenges: API rate limits, DCO compliance, token optimization, and governance rules.
Attendees will leave with:
- A tested framework for evaluating agentic workflows in their own projects
- Battle-tested patterns and pitfalls from production deployments
- Strategies for building hybrid human-agent support communities that share solutions to common blockers
- Actionable recommendations for Linux Foundation to accelerate ecosystem-wide adoption
- Practical next steps to implement these innovations while preserving open source collaboration values
KubeStellar Contribfest
Join the KubeStellar maintainers to make the project better for everyone. You can choose between several opportunities to contribute, and you can count on maintainers from different project areas to help you on your first steps: Our first User Interface, documentation, website, GO lang, and more.
Blue Green and Canary Deployments on AWS EKS using external-dns, Route 53, and KubeStellar
This session explores canary and blue-green deployment strategies using AWS EKS, external-dns, Route 53, and KubeStellar in multi-cluster environments. Based on the speaker's experiences, attendees will gain practical insights into managing clusters across different VPCs and regions.
Canary deployments gradually introduce a new version, while blue-green maintains two identical production environments for extensive testing. The presentation includes detailed information regarding the challenges encountered during deployment on AWS EKS with ALB, based on the author's experiences. Despite following the AWS documentation, hurdles, including those from a lack of EKS knowledge, led to unexpected results. The session presents the potential for using KubeStellar in achieving successful canary and blue-green deployments in multi-cluster environments, avoiding pitfalls encountered along the way.
Edge, Multi-Cluster, and Multi-Cloud Configuration Management with KubeStellar
KubeStellar is a community working on ways to enable the CNCF ecosystem to manage heterogeneous edge locations with precision and without requiring bundling of objects. KubeStellar supports many different kubernetes distributions such as kind, kubernetes, kcp, OpenShift, HyperShift, and others. KubeStellar uses denatured views and logical clusters to transmit and apply standard and customized kubernetes object types to remote edge locations. KubeStellar prioritizes modularity and abstraction to support flexibility, extensibility, and choice to users of our project.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top