KubeCon + CloudNativeCon India 2026

India speaks in 22 languages. Sarvam ensures AI speaks back in every one of those languages, whether in Marathi to a farmer confirming her crops, in Tamil to a citizen accessing a government scheme, or in Hindi to a customer resolving a loan query. Delivering that across millions of API calls and voice conversations daily, at sub-second latency, and while remaining auditable, is as much an infrastructure challenge as an AI one.

In this keynote, we explore how Sarvam uses CNCF technologies to power its conversational AI agent platform and how that platform drove a national crop-verification program that reached 17 million farmers across 4 states in a single 24-day campaign, with a roadmap to 8 crore farmers for the next season. The stack includes Linkerd for a zero-trust service mesh, Strimzi-managed Kafka for event streaming, Argo CD and Helm for GitOps-based delivery across roughly 20 components, Prometheus and Thanos for unified multi-cluster observability, and CloudNativePG as the database of record.

Flipkart, one of India's leading digital commerce entities, runs hundreds of tightly coupled microservices that must withstand the massive traffic surges of Big Billion Days and festive sales. To stop treating outages as an afterthought, Flipkart's Central Reliability Engineering team built a centralized chaos engineering platform on top of Litmus Chaos. To make it work at their scale, they customized Litmus in four meaningful ways: a hybrid multi-tenancy architecture that splits the difference between cluster-wide and namespace-wide installs, a DaemonSet-based high-availability model for chaos injection, a first-class Script Runner fault for dynamic target selection and context chaining, and a hybrid VM chaos extension for workloads that don't run on Kubernetes. This talk shares what they built, why, and what they contributed back upstream.

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.

Two years ago, every platform team was building developer platforms. Today the same teams, and the AI Cloud providers selling to them, face a harder question. How do you safely share GPU infrastructure across multiple AI teams without stepping on each other?

Kubernetes is already the production platform for AI inference. Containers, autoscaling, multi-tenancy, RBAC are all solved. The remaining problem is GPUs themselves. Today Kubernetes asks for them in whole numbers. One pod, one GPU, even when the workload uses 10% of it.

This keynote walks through how the CNCF ecosystem, and Kubernetes itself, has answered. HAMi virtualizes one physical GPU into multiple slices today, each with its own memory budget. DRA evolves the platform's resource model so Kubernetes finally understands GPUs as rich devices instead of opaque numbers.

Then, we move to a live demo. There will be a MacBook on stage, connected directly to an NVIDIA DGX Spark. A single Blackwell GPU will run two open source LLMs for two teams, generating answers simultaneously. There are no slides and no recordings.

That's the AI factory. And yes, Kubernetes has solved it, with a little help from its friends.

National Payments Corporation of India (NPCI) processes more than 24 billion monthly transactions and 700 million daily payments on one of the world’s largest real-time payment systems, all powered by open source, cloud native infrastructure. In this keynote, NPCI will share how they built their AI infrastructure using Kubernetes, Kubeflow, Prometheus, vLLM, OpenStack, NVIDIA AI platforms and many more open-source components to deliver real-time fraud detection in under sub second latency at national scale. The session will explore the architecture behind their payment rail including GPU-accelerated AI inference, distributed cloud infrastructure, and cloud-native tooling powering fraud and AI infrastructure, and financial AI models. Get a look inside their reference architecture and how cloud-native technologies are becoming critical infrastructure for AI at scale with sovereignty.

India moves on Rapido. The country's largest taxi platform serves over 4 million rides daily, matching riders to drivers in real time across 150+ microservices at 200K requests per second.

None of this would exist without the CNCF ecosystem. In this keynote, we walk through how Kubernetes, Istio, Prometheus, Thanos, Cert Manager, and External DNS work together to power a platform where every second of latency means a rider waiting in the heat.

This is the story of what cloud native makes possible when the stakes involve millions of livelihoods, not just uptime metrics.

When tens of millions of fans tune in for live cricket, every millisecond matters. In this keynote, engineers from JioHotstar will take audiences behind the scenes of one of the world’s largest streaming platforms across globally distributed, cloud-native infrastructure.

The session will explore how JioHotstar uses Kubernetes, Amazon EKS, OpenTelemetry, Karpenter, Envoy Gateway, and other CNCF technologies to scale to 50–60 million concurrent users during live sporting events while maintaining reliability, observability, and performance at massive scale. Attendees will see how nearly all core services from personalization and payments to subscriptions and content delivery run in a highly containerized environment spanning dozens of clusters across multiple continents.

The keynote will also preview JioHotstar's next phase of AI-driven operations, including conversational Kubernetes troubleshooting, intelligent SRE automation, and AI-powered user experiences. Through real-world architecture diagrams and live-event scale insights, this session highlights how cloud native infrastructure keeps digital entertainment moving for millions every match day.