This is the home base for a talk-by-talk deep-dive series on KubeCon + CloudNativeCon India 2026, held in Mumbai on June 18–19. This post is the Day 1 field report: what the day felt like, what the keynotes signalled about where cloud-native is heading in India, and a curated index of the 17 sessions I sat in on — each of which gets its own dedicated deep dive (the per-talk posts publish through the series). Day 2 gets its own field report once I've written it up.
KubeCon finally came to Mumbai, and the room felt different from the global editions. India's cloud-native story isn't "how do we adopt Kubernetes" anymore — that fight is largely won. The questions on stage were the harder, second-generation ones: how do we run this at population scale, keep it secure under regulators, pay for the GPUs, and stop drowning in our own platform complexity? Day 1 was a clean cross-section of exactly those themes, and this index is organised around them.
If you only have five minutes, read the keynote recap and the "themes of the day" section below. If you came for the engineering, jump straight to the talk index — every entry links to a full deep dive that unpacks the slides, the architecture, the trade-offs, and the bits worth stealing for your own stack.
How day 1 actually felt
Doors opened early — registration and badge pickup from 8:00am in the Lotus Foyer, coffee flowing, the usual scramble of lanyards and stickers. The opening keynote block kicked off at 9:45am in the main hall (Jasmine 2) and ran as a tight relay of short, punchy talks rather than one long monologue — a format that works well for a kickoff because nobody loses the room.
The energy was unmistakably scale-first. Where a Western KubeCon keynote might open with a developer-experience story, this one opened with sovereignty and population-scale numbers. India runs some of the largest real-time systems on earth — UPI payments, cricket streaming to hundreds of millions, national identity rails — and the speakers leaned into that. The subtext all morning: cloud-native here isn't a hobby, it's load-bearing national infrastructure.
The middle of the day was the usual delightful chaos — Project Pavilion, the Solutions Showcase, lightning talks, hallway-track conversations that were honestly half the value. The afternoon split into parallel technical tracks, which is where the 17 talks below live: a mix of platform engineering, security and policy, the AI/agents wave, observability, storage, and edge.
Fig 1 — the arc of day 1: keynote relay, midday community, four parallel afternoon themes.
The keynotes — what the morning signalled
After the welcome and opening remarks from Jonathan Bryce and Chris Aniszczyk of the Linux Foundation, the keynote relay made the day's thesis obvious. A few that set the tone:
- Sovereign AI (Avi Kothari & Vinayak Gavariya, Sarvam) — building AI capability that a country actually controls, on infrastructure it owns. The "sovereignty" theme recurred all day.
- Chaos as a platform (Aditya Sridasyam, Flipkart & Uma Mukkara, Harness) — chaos engineering productised into a self-serve platform, not a one-off game day.
- LLMs on Kubernetes (Shrinidhi Venkataraman & Nithin R, AstraZeneca) — running large models as first-class Kubernetes workloads, in a regulated pharma context no less.
- AI Factories (Saiyam Pathak, vCluster) — the "factory" framing for standardised, repeatable AI infrastructure.
- Population-scale AI at NPCI (Tittu Varghese) — the organisation behind UPI talking about AI at the scale of a nation's payments.
- Cloud Native at scale (Rapido) and JioHotstar cricket streaming (JioStar, Pradeep Bishnoi) — two of the most extreme real-time scaling stories you'll hear anywhere.
The through-line: AI workloads are now the gravity well the rest of the platform bends around, and scale + sovereignty + cost are the constraints everyone's optimising against. Hold that lens up to the 17 afternoon talks and they snap into focus — almost every one is a different answer to "how do we keep this manageable as it grows."
Themes of the day
Before the index, the four threads that ran through the technical sessions — and which the deep dives are grouped by:
| Theme | The recurring question |
|---|---|
| Platform & app delivery | How do we give hundreds of teams a paved road without every team rebuilding the platform — and without one giant shared cluster becoming a blast radius? |
| Security, policy & identity | How do we get strong guarantees (policy-as-code, least privilege, safe disclosure, federated auth) without slowing delivery to a crawl? |
| AI, agents, GPU & serving | How do we orchestrate agentic systems in production, observe what they actually did, serve models that scale, and not set GPU budget on fire? |
| Observability, edge & storage | How do we see what's happening (cheaply), push compute to the edge, and run stateful storage on Kubernetes without fear? |
The deep-dive index — 17 talks
Every talk below gets a dedicated post that walks the slides end to end: the problem, the architecture, the demo, the gotchas, and what's worth adopting. Links go live as each deep dive publishes through the series.
// Platform engineering & app delivery
Re-Architecting Monoliths into Kubernetes Microservices
A practical migration playbook — how to peel a monolith into services on Kubernetes without a doomed big-bang rewrite, and what actually breaks along the way.
Aditya Sharma DEEP DIVE 02Shared-First Kubernetes Platforms
The case for designing platforms as shared-by-default — multi-tenancy, paved roads, and where the line sits between "shared" and "isolated."
Platform engineering track DEEP DIVE 03When Kafka Goes Cloud Native — Observability That Actually Works
Making Kafka-on-Kubernetes truly observable with a vendor-agnostic OpenTelemetry pipeline — the three signals, reading ISR, and the metric-cardinality trap.
Mary Vinothini S · Roopadharsini K · Fidelity Investments DEEP DIVE 04KubeVela — One Interface To Rule Them All
The Open Application Model in practice: separating app developers from platform plumbing with KubeVela's abstractions, Crossplane, and one manifest for apps + cloud + AI.
Gowtham S · Jerrin Francis · Guidewire Software DEEP DIVE 05Beyond the Primary CR — Advanced Operator Patterns
What happens after the hello-world operator: secondary resources, ownership, status, and the patterns that keep complex controllers sane.
Guna K Kambalimath · Kishen V · IBM// Security, policy & identity
The Kyverno Five — Policy-as-Code Patterns
Five high-leverage Kyverno policies every cluster should run, and the mental model for writing policy that protects without blocking delivery.
Neha Jaju · Nirmata DEEP DIVE 07Root Without Risk — Kubernetes User Namespaces (KEP-127)
How hostUsers: false remaps container root to an unprivileged host UID — the idmap-mount breakthrough and the nine-year road to GA in v1.36.
Commit-Then-Disclose — Verifiable, Private SBOMs
A SHA-256 commitment scheme that proves an SBOM is untampered while keeping components sealed — selective disclosure to auditors in 481-byte proofs, no ZKP needed.
merkleSBOM · supply-chain security track DEEP DIVE 09Keycloak Federated Client Authentication
Secretless OAuth clients: Keycloak 26.6 validates Kubernetes service-account and SPIFFE workload tokens instead of storing client secrets.
Rishabh Singh · Red Hat// AI, agents, GPU & serving
Building & Orchestrating Production-Ready Agentic AI Systems
From one LLM call to a production agentic system with LangChain4j — the four orchestration patterns, the AgenticScope, and the autonomous supervisor pattern.
Kevin Dubois · Daniel Oh · IBM / Red Hat DEEP DIVE 11What Did My Agent Do? — Observability for AI Agents
You can't debug what you can't see: traces and spans, single- vs multi-turn primitives, OTel GenAI conventions, zero-code eBPF instrumentation, and evaluating reasoning.
Ishan Jain · Grafana Labs DEEP DIVE 12Zero-GPU Autopilot — Kagent & Kgateway for Self-Healing Clusters
A self-healing cluster autopilot with no GPUs and no external APIs: CPU-only vLLM powering Kagent (reasoning) and Kgateway (the Gateway API guardrail), all inside the VPC.
Ashok M · DigitalOcean & Dillibabu Sampath · Wells Fargo DEEP DIVE 13Plug In and Scale — Serving LLM Models on Kubernetes
Inference vs serving, the runtimes (vLLM/SGLang/Triton), disaggregated serving, the vLLM Production Stack, sleep mode, KEDA scale-to-zero, and AIPerf benchmarking.
Shrinidhi Venkataraman · Nithin Rajan · AstraZeneca// Observability, edge & storage
The Lean Observability Stack — Native Telemetry for Service Mesh
Golden signals, alerts, and SLOs for free from Istio's in-Envoy stats filter — labels, templateable SLOs, cardinality control, and Envoy logs via OTLP.
Arpitha Malavalli · Google DEEP DIVE 15KubeEdge Deep Dive
Extending Kubernetes to the edge with KubeEdge — architecture, the cloud/edge split, device management, and running workloads where the data is.
Ronak Raj · CNCF KubeEdge DEEP DIVE 16Rook — Cloud-Native Storage, from the Maintainers
Running Ceph storage on Kubernetes with Rook: architecture, operations, and the hard-won maintainer wisdom for stateful workloads.
Rook maintainers DEEP DIVE 17Bonus session deep dive
One more talk from the day — the full breakdown lands with the rest of the series. (Title finalised in the deep dive.)
to be detailedDay 1 takeaways
- AI is the new gravity. Sovereignty, GPUs, serving, and agents dominated the agenda — the platform exists to feed the models now.
- Scale is assumed. The interesting questions weren't "can Kubernetes do this" but "how do we do it at population scale, under regulators, without burning the budget."
- Platform complexity is the quiet villain. Shared-first platforms, app-delivery abstractions, lean observability, and "root without risk" are all attacks on the same enemy: accidental complexity.
- Security shifted left and got practical. Policy-as-code, federated identity, and coordinated disclosure showed up as everyday engineering, not a separate silo.
- India's cloud-native scene has its own accent. Real-time, huge, sovereign, cost-sensitive — and increasingly setting the agenda rather than following it.
FAQ
Is this the official schedule or your notes?
It's my field report and reading of the talks I attended, anchored to the published schedule. The keynote line-up follows the official agenda; the deep dives are my own expanded write-ups of each session's slides, not official transcripts.
When do the per-talk deep dives publish?
They roll out through the series — each links from the index above as it goes live. Day 2's field report and its deep dives follow once I've written them up.
Why group by theme instead of by time slot?
Because the value is in the ideas, not the clock. Grouping by theme makes the connections obvious — all the platform talks rhyme with each other, as do the security and AI ones.
References
- KubeCon + CloudNativeCon India 2026 — full schedule · sched.com
- KubeCon + CloudNativeCon India · event home (Linux Foundation)
- CNCF · the projects behind most of these talks