DevOps Midwest — 2026 Speakers

Manual server rotation during deployments creates structural failure modes that checklists cannot prevent. When distributed teams operate across time zones, the architectural separation between load balancer management and CI/CD execution causes recurring outages regardless of how careful engineers are.

This talk presents a reference architecture that eliminates these failure modes by design,not by procedure. By treating load balancer pool membership as a first-class CI/CD concern and integrating F5 API controls as native Jenkins pipeline stages, we eliminate three structural vulnerability vectors completely.

Sai Joshitha Kathari is a Senior Site Reliability Engineer in the Fintech Industry based in Dallas, Texas. She specializes in high availability, reliability, observability, and resilience of large-scale distributed systems. Her work focuses on identifying and eliminating structural failure modes in production environments, ensuring service continuity for high-revenue enterprise clients operating across multiple data centers and time zones.

Delve into the intricacies of Kubernetes deployment strategies in this informative talk. I'll provide a clear breakdown of various strategies and guide you on when to employ each one effectively. The highlight of the session is a live demo, offering a hands-on exploration of the inner workings behind each deployment strategy.

From rolling updates to blue-green deployments, we'll navigate the landscape of Kubernetes deployment options, emphasizing practical insights for diverse scenarios. The live demo aims to demystify the deployment process, giving you a solid understanding of the mechanics at play.

Whether you're new to Kubernetes or a seasoned user, this talk equips you with actionable knowledge to enhance your deployment strategies. Join us for a practical journey through Kubernetes deployments, gaining valuable insights to optimize your development workflow.

Many engineering organizations want better insight into how their software delivery systems are performing. Leaders ask questions like:

• How long does it really take us to deliver software?
• Where are we getting stuck in the delivery pipeline?
• Are we improving over time?

Teams often assume answering these questions requires new tools, specialized analytics platforms, or complex reporting systems.

In reality, most teams already have the data they need.

Issue trackers, CI/CD pipelines, and version control systems capture a large amount of delivery information. The challenge is not collecting the data. The challenge is understanding which metrics matter and how to extract them.

In this session we’ll explore how to measure key delivery indicators using the data that already exists inside common development tools. We'll look at practical examples using issue tracking systems and build pipelines to calculate metrics such as cycle time, deployment frequency, and delivery throughput.

Attendees will leave with a practical approach for building meaningful delivery metrics without introducing new tools or heavy reporting overhead.

Every company with a software team faces the same pressure to ship faster without breaking things. When your company has real security concerns and skeptical leadership, "just turn on an AI coding tool and see what happens" isn't a strategy.

This session walks through a structured, phased pilot of OpenAI Codex at a company whose software team builds and maintains the core system that runs the business. The specific use case: generating the mountain of missing unit tests never written over years of moving fast. The pilot was part of a deliberate shift-left quality initiative -- and eight weeks in, cycle time on test writing had dropped by more than 40%.

You'll see how we built the ROI case for skeptical leadership, how we addressed security and compliance requirements before a single developer touched the tool, and why we chose unit tests specifically to manage code security and quality risks. We'll also cover how we designed the pilot phases to generate visible evidence of responsible governance: documented guardrails, weekly retrospective summaries, and team-developed prompting best practices.

Attendees will leave with a reusable pilot framework applicable whether they're a developer wondering if AI tools are actually worth it, a manager trying to build the internal case for adoption, or a technical leader trying to navigate the governance conversation without killing the initiative.

For years, our delivery pipelines and testing strategies have been built on a simple assumption: system behavior is predictable. That assumption breaks the moment AI enters your architecture.

AI-powered services don’t behave like traditional components. They produce non-deterministic outputs that look consistent - until they aren’t. And when that variability flows through automated validation and release processes, it introduces a new class of risk that traditional approaches aren’t designed to catch.

So how do you maintain confidence in releases when “correct” is no longer binary?

In this session, we’ll break down how non-deterministic AI behavior impacts modern delivery, and expose the common traps teams fall into when treating AI like conventional software. We’ll walk through practical techniques like similarity-based validation, structured input variation, and LLM-based evaluation that allow you to assess AI behavior in a repeatable, automated way.

You’ll leave with a clear, practical approach to adapting your testing and release strategies for AI-enabled systems so your pipelines aren’t just fast, but trustworthy.

This presentation explores the evolution from traditional software development to AI-first development workflows, backed by comprehensive performance data and real-world implementation results.

Drawing from extensive industry research and hands-on AI system implementation experience, this session will demonstrate how organizations achieve remarkable improvements in model development velocity, system reliability, and AI application performance. Attendees will discover how containerized ML pipelines with Docker dramatically reduce model deployment time while Kubernetes orchestration maintains sub-100ms inference response times even at 95th percentile measurements across distributed AI workloads serving millions of requests daily.

The presentation covers critical AI development tools including MLOps platforms that significantly reduce model time-to-production, API-first ML serving architectures that accelerate integration cycles, and event-driven AI systems processing up to 500,000 inference requests per second. Advanced monitoring and observability practices will be explored, showing how distributed model tracking cuts debugging time while centralized experiment logging improves model performance correlation substantially.

Real case studies demonstrate tangible business impact: a global financial institution achieved 4x faster fraud detection model deployment and 60% reduction in false positives, while a major retailer reached 99.9% AI service availability during peak traffic, contributing to $85 million in additional annual revenue through personalized recommendations. Security implementations show 70% reductions in model vulnerability exposure and data pipeline incidents.

Key takeaways include proven AI governance frameworks that substantially improve cross-team collaboration on ML projects, domain-driven AI architecture principles that reduce model coupling, and hybrid cloud strategies enabling 95% higher AI deployment success rates. Attendees will leave with actionable insights for implementing AI development toolchains that deliver measurable ROI while positioning organizations for future growth in an increasingly AI-driven technology landscape.

There is no shortage of AI hype in DevOps. There is also no shortage of teams who tried copilots and came away unimpressed. This talk is for them.

We skip the magic demos. Instead, we look at what actually shifts when AI agents work as persistent collaborators instead of autocomplete on steroids. Where agentic workflows produce measurable gains today. Where they do not. And how to tell whether AI is accelerating your cycle time or adding a new failure mode to your on-call rotation.

If you are skeptical, good. This session was built for that.

The path from “I have a model” to “it’s serving production traffic on my cluster” is full of sharp edges. This talk is an end-to-end walk-through: GPU device plugins and node scheduling, model weight distribution strategies for 10–140 GB artifacts, health checks that survive multi-minute model loads, and GPU-aware auto-scaling. Includes a live demo deploying an open-weight LLM on Kubernetes and serving inference traffic with real-time GPU metrics.

Rohit works as a cloud solutions architect & customer engineer at Google Cloud, focusing on core cloud platform capabilities (compute, network, cloud security, storage & AI infrastructure) with a specialization on containerized workloads & container orchestration (Kubernetes). He guides some of Google Cloud's largest customers from retail, manufacturing, and food/restaurant industries towards successful migrations, modernization, and transformation journeys, providing them guidance on continuous improvement, operational efficiencies & adoption of AI at scale with Google Cloud.

Staying current with agentic coding best practices can feel like a second job, and agent skills/plugins for cloud operations are no exception. Maybe you already automate cloud tasks with agents, but your feed is full of starred skill examples you never quite revisit—so you keep solving problems others have already solved. Or maybe you're fluent in agentic coding but haven't yet applied it to cloud ops. Either way, it's hard to know where to begin.

In this session, I break down seven open source agent skills and skill patterns for public cloud environments. I cover workflows for cloud operations, including IaC management and drift detection, policy as code, and integrating open source tools into complex agent workflows. Expect actual code—real SKILL.md files—and a clear grasp of the broader patterns for building agentic workflows for public cloud environments. Whether you've built agent skills before or are just starting out, you'll leave with concrete examples you can put to work.

Tara Schofield is a Technical Advocate at Datadog and an AWS Community Builder. Before Datadog, she worked at AWS as a Cloud Computing Consultant, building and delivering cloud solutions for security-sensitive customers.

Modern software teams face a persistent challenge in balancing rapid delivery with system reliability. While DevOps practices have accelerated deployment cycles, integrating compliance and governance requirements often introduces friction that slows delivery and increases operational risk. This tension becomes more visible in environments where stability and traceability are just as critical as speed.

This session presents a compliance native CI CD architecture that embeds governance directly into pipeline execution, enabling teams to move quickly without sacrificing reliability. Instead of treating compliance as a final validation step, the approach integrates policy enforcement, artifact integrity verification, and automated audit generation into each stage of the delivery lifecycle. This ensures that every deployment meets predefined standards while maintaining continuous flow.

The architecture introduces automated policy gates that evaluate code quality, security posture, and configuration compliance in real time. Immutable artifact tracking ensures traceability from code commit to deployment, while modular pipeline design allows teams to scale workflows across multiple services without increasing complexity. By shifting compliance into the pipeline, manual interventions are minimized, reducing bottlenecks and human error.

A key focus is achieving operational stability alongside high velocity. The framework supports controlled release strategies, environment promotion workflows, and rollback mechanisms that enhance resilience without slowing down teams. Automated reporting capabilities provide visibility into system health and compliance status, allowing teams to maintain confidence in every release.

This approach demonstrates how integrating governance into DevOps pipelines enables organizations to achieve both speed and stability. Attendees will gain practical insights into designing scalable CI CD systems that support rapid innovation while ensuring consistent, reliable software delivery.

In this live coding session, Joel will demonstrate four techniques for better testing:

1. leveraging wiremock for API dependencies
2. acceptance testing for legacy code
3. refactoring for readability (of course)
4. in memory / container based databases for DB dependencies (mainframe, db2, etc)

Modern software is no longer built from code your team writes alone. It is assembled from open-source packages, containers, SDKs, SaaS integrations, generated code, and third-party services layered across the stack. In many applications, the majority of the code running in production was never authored, reviewed, or maintained by your developers. That means much of your attack surface lives outside your repo.

This session explores why dependency risk has become one of the biggest engineering and security challenges in cloud-native development. Using incidents such as Codecov, Log4Shell, SolarWinds, and the XZ Utils backdoor, we examine how a single compromised component can move silently through CI/CD pipelines, artifact registries, containers, and production environments.

Rather than focusing only on CVE counts, this talk shows developers how supply chain risk actually enters systems: transitive dependencies, over-permissioned packages, abandoned libraries, tampered build pipelines, unsigned artifacts, and blind trust in upstream maintainers. We will break down why traditional vulnerability scanning alone often creates noise without reducing real exposure.

Attendees will learn practical engineering controls that work: Software Bills of Materials (SBOMs), dependency minimization, version pinning, signature verification, provenance attestation, reproducible builds, least-functionality package selection, runtime inventory, and policy gates in CI/CD pipelines. We will also discuss how AI-generated code can introduce risky packages faster if dependency hygiene is weak.

Developers will leave with actionable patterns to reduce blast radius, improve visibility, and build software supply chains that are faster, safer, and easier to trust.

Infrastructure-as-Code reviews often miss risk because reviewers see individual pull requests without the architecture context needed to understand blast radius, ownership, historical patterns, and cross-resource dependencies. This technical talk introduces SecReviewAgent, a context-aware DevSecOps review pattern for Terraform, Kubernetes, and cloud infrastructure changes.

The session shows how to combine deterministic scanners, dependency mapping, architecture memory, and LLM-assisted reasoning while keeping the model bounded: scanners and policies provide enforcement, the AI helps summarize risk and explain tradeoffs, and humans retain approval authority.

Attendees will see a practical workflow for prioritizing high-risk changes, reducing repetitive review comments, protecting sensitive repository details, and building safer CI/CD review gates for infrastructure changes.