KubeCon + CloudNativeCon India 2026

Our session introduces a cloud-native, in-house platform for modular LLM inference at enterprise scale. Built on Kubernetes, the architecture unifies open-source and vendor models via OpenAI-compatible APIs and supports distributed serving with popular inference runtimes like vLLM, SGLang, and Triton.

Powered by NVIDIA Dynamo, the system optimizes GPU fleets through intelligent scheduling, KV-aware routing, prefix caching, and NIXL-based GPU-to-GPU data transfer, further optimized by allocating fractions of a GPU using KAI-Scheduler. The platform delivers streaming, speculative decoding, quantization, and autoscaling to zero via KEDA.

We ensure comprehensive observability with Prometheus, Grafana, and ELK, all governed by GitOps principles using ArgoCD and secured with enterprise-grade practices. For end-user consumption, the platform integrates with Open WebUI via standard APIs. We’ll cover the architecture, key components, and cloudability-driven cost governance strategies that empower data science teams while accelerating safe, sustainable AI innovation across AstraZeneca.
Our session introduces a cloud-native, in-house platform for modular LLM inference at enterprise scale. Built on Kubernetes, the architecture unifies open-source and vendor models via OpenAI-compatible APIs and supports distributed serving with popular inference runtimes like vLLM, SGLang, and Triton.

Peer Group Mentoring allows participants to meet with experienced open source veterans across many CNCF projects. Mentees are paired with 2 – 10 other people in a pod-like setting to explore technical, community, career, and certification questions together.

If you're interested in attending as a Mentee, seats are only guaranteed to those on the sign up sheet.

If you're interested in being a Mentor, please sign up by May 25.


Check-in: 11:30 AM - 11:40 AM
Session: 11:45 AM - 1:00 PM

The Indian BFSI sector is shifting to cloud-native architectures to enhance operational efficiency, scalability, and digital customer experiences. This panel of industry leaders, Architects from India's largest Bank (SBI) to leading Banking software firms (Infosys Finacle, TCS, I-exceed), is here in India's financial capital, to discuss:
- Cloud Native adoption and growth in the BFSI sector
Banking-specific, unique real-world use cases, role of Agentic AI
- Developing and operating applications at scale, e.g 1.7B accounts serving 1.3B people across the globe
- Security safeguards, governance, and change management, with the use of the CNCF project Flagger
- Tackling challenges with cloud-native technologies and the feedback loop with the community and maintainers.

This panel shares insights on cloud-native adoption in India's BFSI sector, incorporating global perspectives. Be inspired by how popular technologies are architected at scale to deliver innovative banking solutions.

We first trace the evolution of the "aerodynamics" of the K8s data, control, and work planes. As we dissect the architectural building blocks and controller archetypes of K8s, we discover the kubelet controller pattern as an universal design principle, with which we easily explain the unfolding of K8s Inception with Hosted Control Planes (HCP), where any control plane becomes a recursive workload in K8s itself.

Armed with these concepts we come to the ultimate, supersonic plane: KCP, the pure data and API plane.
Learn how to mentally decouple K8s from its underlying infrastructure and move toward a truly scalable, declarative API-first future. The next evolution in cloud-native. With KCP, and by treating KRM as a universal interface, we will show how you can benefit from the same pattern used to manage a single pod, to engineer global-scale platforms.

At each architecture evolution step, we accompany a demo and use case: from single clusters, vClusters, clusters with HCP, and KCP.

When timelines are tight or infrastructure is isolated, setting up complex observability suites can be a bottleneck. However, a service mesh like Istio, backed by Envoy, already provides a goldmine of data. Drawing from experience in restricted, air-gapped environments, we demonstrate how to turn Envoy and Istio into primary telemetry sources to build a reliable pipeline using Prometheus, Grafana, and OpenTelemetry (OTel). We will break down essential Istio and Envoy metrics—such as request totals, bytes, and duration—and demystify their rich labels. Attendees will learn to leverage these metrics for "Golden Signal" dashboards, SLOs, and meaningful alerts, without manual per-service configuration. We also cover setting up OTel-based pipelines for Envoy access logs. Finally, we address "day 2" operations: using Prometheus relabeling to reduce cardinality, managing log verbosity via the Istio Telemetry API, and determining when to use synthetic probers over standard metrics.

A match made in heaven, or one hell of a couple? Platform teams spent 2 years perfecting golden paths for cloud-native infrastructure. 5 engineers managing 100 services.

Last week, a new AI agent generated the same config in 30 seconds. This week they're managing 5 LLM models across 200 services. The future seems one-sided. This couple needs counselling.

This is escalating abstraction: Platforms reduce k8s chaos. AI automates those paths at breakneck speed. But AI-generated configs violate platform policies. Platforms enforce governance via admission controllers. Governance needs observability. Observability reveals the gap: no model lifecycle management, no agent audit trails, no cost control APIs.

This talk traces real interdependencies: deprecating models, managing sprawl, isolating agents, controlling budgets. You'll learn what exists - CNCF AI Conformance, tools, working groups & the critical gaps.

Right now, there's no greater agony than rationalising AI without platforms.

Telemetry travels long pipelines before reaching observability backends. While enrichment, filtering, and sampling provide clear diagnostic value, much of the compute cost comes from repeatedly converting telemetry between different data formats.

Telemetry flows through SDK representations, wire protocols, collector-internal formats, and backend ingestion schemas. Each boundary introduces marshaling, unmarshaling, and copying. These transformations add no new information, yet consume CPU and memory and scale linearly with data volume—creating a hidden “transform tax” that compounds at scale.

This talk presents measurements from instrumented OpenTelemetry SDK and Collector pipelines, quantifying compute spent on pure format conversion versus value-generating processing. Attendees will learn where conversion costs arise and explore strategies to reduce waste, including fewer representation hops, zero-copy techniques, and emerging approaches such as Apache Arrow-based layouts.

Kubernetes makes scaling workloads easy, but as clusters span regions and clouds, understanding and trusting costs becomes difficult. Limited cost transparency in platform challenges customer trust, making cost visibility a platform responsibility with accuracy, transparency, and trust as first-class requirements.

In this talk, we share how we centralised our platform’s k8s cost metrics using OpenCost, VictoriaMetrics, and Grafana to deliver consistent, customer-facing visibility across multi-region, multi-cloud clusters.

We will walk through the architecture, show how raw metrics are converted into application-level cost data, and how sharing billing-level cost data improved transparency, trust, with 99.9% accuracy. The session also covers production lessons, trade-offs, and operational challenges while scaling centralised cost visibility and customer trust across the platform.

In this panel, engineering leaders from Intuit and Atlassian discuss how they transitioned from fragmented monitoring to a unified, OTel-based observability strategy across 1000+ microservices, built on CNCF standards like OpenTelemetry and Argo.

Key discussion points include:
- Standardization at Scale: How to drive OpenTelemetry adoption across thousands of microservices without slowing down feature delivery.
- The Cost of Insight: How to move beyond the "collect everything" mentality. We’ll discuss the architectural shift from mindless ingestion to intelligent sampling
- AIOps & The Future: Moving beyond dashboards to predictive incident response and AI-augmented root cause analysis.
- 'Signal' vs. 'Noise':How are teams leveraging AI to separate true signals from the noise of thousands of clusters?

Join us for a candid conversation on the technical hurdles, cultural shifts, and architectural "oops" moments encountered while scaling observability for millions of customers.

A common challenge for SaaS platforms on K8s is handling the full tenant lifecycle, onboarding, upgrades, and teardown, without manual provisioning or tenant-per-cluster architectures. As tenant counts rise, manual workflows slow customer time-to-value and create operational toil.

This session shows how we built an event-driven tenant lifecycle engine using Backstage as the self-service entry point and Argo Workflows, argo-events, NATS, and GitOps as the automation backbone. It handles onboarding, provisioning, feature enablement, scaling, migrations, and teardown without human intervention. The platform reduced onboarding from hours to minutes and supported ~10,000 tenant-specific deployments across clusters while avoiding cluster-per-tenant explosion.

Attendees will learn how Backstage can drive self-service provisioning flows while Argo Workflows and GitOps execute lifecycle operations reliably, with patterns that reduce toil and improve reliability and SLO compliance.

Your release pipeline fails, and you ask an AI agent to debug it. You paste logs, grant doc access, and it fixes the bug. But where did that prompt actually go?

Every time you interact with AI, you create a complex data flow. A single request can pull from a vector db (RAG), trigger external tool calls via MCP, & leave sensitive trails in app logs. Now, the most sensitive data in any org is no longer in databases. It's in prompts, logs, & agentic calls.

In 2025, half of employees have leaked sensitive company data into public AI tools. When these prompts cross borders, sovereignty is lost.

In this talk, I'll demo how to generate an “AI Receipt”, a per-request trace using open source tools (OTel). Tracing the full data path - prompt ingestion, retrieval, & tool calls, see exactly where sovereignty breaks, turning sovereignty from a claim into verifiable proof of where data flows.

Join me to create your AI Receipt - so next time you prompt, you know exactly where it goes.

GitOps has revolutionized the way we manage cloud-native infrastructure but what happens when your workloads need to be confidential, attested, and cryptographically verifiable?

This session explores a new frontier in GitOps: deploying and managing confidential container workloads with Argo CD, Sigstore, and Confidential Computing hardware (e.g., Intel TDX, AMD SEV, IBM Secure Execution). You’ll learn how to enforce cryptographic integrity, verify runtime isolation, and prevent unauthorized workloads using policy-driven rollouts and attestation checks.
We’ll showcase how to:
• Automate confidential workload verification in Argo CD pipelines
• Use Sigstore/Cosign for keyless signing and provenance attestation
• Gate deployments based on SLSA levels or TEE runtime evidence
• Visualize runtime trust using OpenTelemetry and Prometheus

As a Kubernetes maintainer and ex-CNCF TOC Vice Chair, I’ve been involved in leadership decisions at both ends: stepping into roles and intentionally stepping away from them.

In recent years, Indian contributors have stepped up into maintainer & governance roles. This is a huge win. But the next challenge is quieter and far more uncomfortable: how do today’s maintainers step aside to make room for the next generation of leaders?

Too often, maintainer succession happens only in crisis - after burnout or sudden departures. Projects stall, contributor momentum drops & leadership remains concentrated in too few hands.

In this lightning talk, I’ll break down what effective succession looks like in practice, drawing from real examples from K8s and CNCF projects. I’ll cover maintainer tenure and transition patterns, real succession stories, emeritus maintainer models, psychological challenge of letting go as well as succession templates, role definitions and knowledge transfer checklists.