Cloud Native Day Pune 2026 (Invite Only)

Security is an outlier in the cloud native world. Some folks may take offense at this, but that is the unfortunate truth. In this session, I aim to educate cloud native engineers on the topic of security-first development, especially in an open source world. Software supply chain attacks are at an all-time high. Defaults are not secure enough in the containers world. Lots of ongoing work within the Security-TAG of the CNCF will be highlighted during this talk. Special focus will also be placed on the work of the OpenSSF (Open Source Security Foundation). Let's make our cloud native footprints safer!

What if your AI didn't just answer questions, it delegated work, ran tasks in parallel, and delivered results across your entire toolstack? In this workshop, we'll explore two powerful open-source AI agent frameworks: OpenClaw and Hermes Agent, both deployed on Vultr. We'll observe live interactions through messaging platforms, and orchestrate real multi-agent workflows where specialized AI agents collaborate autonomously, all powered by Vultr Serverless Inference models. Vultr acts as the backbone, providing the compute and serverless inference infrastructure that makes self-hosted AI agents fast, affordable, and production-ready. From building multi-agent pipelines that run parallel research tasks to delivering results across platforms in real time, this workshop gives you a front-row seat to the future of AI-powered automation.

As organizations move from simple LLM applications to large-scale GenAI and Agentic AI systems, infrastructure costs are increasing rapidly. Continuous inference, autonomous agents, multi-model workflows, and dynamic runtime behavior are creating new challenges in scalability, GPU utilization, latency, and operational efficiency.

This talk explores how Kubernetes can become an inference-aware platform for modern AI workloads. We will discuss practical cloud-native strategies for optimizing GenAI and Agentic AI systems using intelligent autoscaling, workload-aware scheduling, dynamic batching, efficient GPU utilization, and scalable model serving patterns.

The session focuses on real-world architectural approaches for building cost-efficient and scalable AI infrastructure without being locked into a specific vendor ecosystem. Attendees will also learn how Kubernetes-native and open-source technologies can help improve resource efficiency, reduce inference cost, and support production-ready AI platforms for the next generation of intelligent applications.

Whether you work in AI infrastructure, platform engineering, or MLOps, this session provides practical insights into running GenAI and Agentic AI workloads efficiently at scale.

Shipping AI was the easy part. Operating it is the job nobody warned you about. Every team is suddenly running AI inference in production, and most are bleeding money. The Kubernetes patterns that served stateless web apps for a decade quietly break when the workload is a 70B-parameter model. This talk is for the platform engineers, SREs, and cloud-native operators now responsible for keeping AI workloads up without a blank cheque.

In 30 minutes, we'll cover:

• Why inference workloads behave nothing like the web services your cluster was designed for
• 5 levers you actually control
• The CNCF-native stack that makes it portable across any cloud (KEDA, OCI, OpenTelemetry etc.)
• The one part of your stack you can't autoscale, and why it matters more than the GPU Participants will walk out with a checklist you can run against your own cluster on Monday morning.