CNCF x AAIF presents: Agentic AI Night: AI Infrastructure
Speak at CNCF Sydney
We run regular meetups for Sydney’s cloud native community, and we’re always looking for people with something real to share. If you’ve built it, broken it, scaled it or had to fix it at 2am, we’d like to hear about it.
What we’re looking for
Talks grounded in practice rather than theory. Kubernetes, platform engineering, observability, security and supply chain, GitOps, service mesh, AI and ML infrastructure, developer experience, and anything else in the cloud native ecosystem. War stories are especially welcome. What broke, why, and what you’d do differently is usually the most useful talk in the room.
Formats
Lightning talk, 5 to 10 minutes. Or a full talk, 20 to 25 minutes plus questions.
A note on vendor content
As a CNCF chapter we keep talks focused on open source and practical engineering rather than product walkthroughs. Referencing your own tooling where it’s genuinely part of the story is completely fine, and speakers do it all the time. Our audience is there for how something works and what you learned building it, so talks framed that way land much better.
First time speaking?
Please submit anyway. A good chunk of our speakers are giving their first talk, and we’ll help you shape the abstract and run through the material beforehand if that’s useful. Everyone starts somewhere.
What we need from you
A title, a short abstract covering what you’ll talk about and what people will take away, and a line or two about yourself. That’s it.
What happens next
We review submissions as they come in and will get back to you either way. If we can’t fit your talk into the next meetup, we’ll keep it on file for the following one rather than making you resubmit.
Submit a proposal
Join us for an evening of insightful talks, valuable connections, and great company.
This one is a collaboration: CNCF Sydney x AAIF are bringing the cloud native and agentic AI communities into the same room.
Everyone is shipping agents. Far fewer people are talking about what it takes to actually run them. As agentic AI moves from prototypes to production, the infrastructure underneath matters as much as the models and the agents themselves.
Speaker: Hrittik Roy, Founding Platform Advocate at vCluster and CNCF Ambassador Talk 1: From Local Inference to Kubernetes: What Changes When Your LLM Gets Busy? Details: Running an LLM locally is a great starting point. But what happens when one user becomes hundreds, requests compete for GPU memory, and response times start climbing?
This talk follows an inference request from prompt to generated tokens, explaining prefill, decode and the KV cache. We’ll explore how batching, model caching and GPU scheduling affect latency and throughput, then introduce disaggregated inference: running prefill and decode on separate workers, including when the added complexity actually makes sense. Using vLLM and NVIDIA Dynamo as examples, we’ll connect these concepts to serving models on Kubernetes and measuring what matters: time to first token, token generation speed, and GPU utilisation.
You’ll leave understanding what happens inside an LLM inference request, where the bottlenecks appear as you move beyond local serving, when disaggregated inference is worth considering, and the Kubernetes building blocks for GPU workloads.
Speaker: Olga Mirensky, Senior Staff Site Reliability Engineer at Ping Identity Talk 2: Platform Engineering for AI Agents Details: Managed agent platforms give you runtimes, sandboxes and enforcement points, but not an internal platform. Drawing on enterprise delivery in financial services and tech, this talk walks through a multi-agent architecture on a cloud managed platform, and the layer you still build yourself: policies derived from your own architecture, reusable blueprints, and telemetry across a fleet of agents.
Platforms make agents fast to build. Deciding what’s worth building is still the hard part.
On the night: Free entry · Food & drinks · Networking · Live Q&A
Thanks to Google for hosting us, and to the AAIF team for partnering on this one.
Seats are limited - RSVP to save your spot. See you there! 👋
-
6:00 PM - 6:45 PM AEDT
From Local Inference to Kubernetes: What Changes When Your LLM Gets Busy?
in-personRunning an LLM locally is a great starting point. But what happens when one user becomes hundreds, requests compete for GPU memory, and response times start climbing? This talk follows an inference request from prompt to generated tokens, explaining prefill, decode and the KV cache. We’ll explore how batching, model caching and GPU scheduling affect latency and throughput, then introduce disaggregated inference: running prefill and decode on separate workers, including when the added complexity actually makes sense. Using vLLM and NVIDIA Dynamo as examples, we’ll connect these concepts to serving models on Kubernetes and measuring what matters: time to first token, token generation speed, and GPU utilisation. You’ll leave understanding what happens inside an LLM inference request, where the bottlenecks appear as you move beyond local serving, when disaggregated inference is worth considering, and the Kubernetes building blocks for GPU workloads.
SPEAKERS -
6:45 PM - 7:30 PM AEDT
Platform Engineering for AI Agents
in-personManaged agent platforms give you runtimes, sandboxes and enforcement points, but not an internal platform. Drawing on enterprise delivery in financial services and tech, this talk walks through a multi-agent architecture on a cloud managed platform, and the layer you still build yourself: policies derived from your own architecture, reusable blueprints, and telemetry across a fleet of agents.
Platforms make agents fast to build. Deciding what’s worth building is still the hard part.
SPEAKERS
No sessions match the selected labels.