2025-09-24

🎉 Join us for the Melbourne AWS User Group Meetup on Wednesday 24 September! 🎉

As usual, we’re back on the last Wednesday of the month in the NAB Events Hub on 395 Bourke and spring has arrived! We’ve got warm days! And days when it’s cold! And they’re usually right after the other! So we know we’re still in Melbourne. Anyway, let’s huddle up together, talk about our hayfever, and enjoy some technical talks.

We have two talks lined up for this month, Gokul will take us through some ways to make Kubernetes run better on AWS, while Renaldi will be talking about a GenAI blueprint for Bedrock. And of course we also have our regular sections, covering what’s going on in the community and what new. goodies AWS has brought us in the past month.

As a reminder, we will start with the food and socialising at 6 pm, but the actual contents of the meetup (and live stream) will begin at 6:30.

Our thanks to Mantel Group for their continued sponsorship.

See below for the full agenda. The event will be streamed on YouTube as always.

Agenda

Enhance Kubernetes cluster performance and optimise costs

• Gokul Venugopal

Level 300

In this session we’ll look at the path of scaling Kubernetes on AWS and how the available tools have evolved:

1. Cluster Autoscaler – an introduction to how it works and the limitations you’ll run into.
2. Karpenter – what it offers compared to Cluster Autoscaler and what’s involved in migrating.
3. EKS Auto Mode – the next step forward, with its advantages, trade-offs, and migration details.

Along the way we’ll discuss the pros and cons at each stage, how to migrate without too much friction, and the common gotchas you’ll want to avoid.

The Policy-Gated Bedrock Blueprint: A Zero-ETL, Guardrailed, Multi-Tenant GenAI Architecture on AWS

• Renaldi Gondosubroto

Level 300

This talk will walk through designing a production-grade, multi-tenant GenAI platform where policy is a first-class control plane (IAM + AWS Verified Permissions) that gates which models, tools, knowledge sources, and token budgets any user or tenant may access. The data plane uses Zero-ETL patterns and Bedrock Knowledge Bases to avoid brittle pipelines while preserving data residency. Orchestration with Step Functions and Lambda implements prompt chaining and tool use, with Bedrock Guardrails for safety and grounding. Finally, we embed observability and cost control and show variants for serverless, containers, and when to fold in SageMaker endpoints for specialized inference or fine-tuning. You’ll leave with a reference blueprint you can deploy, extend, and defend in an architecture review.

We are always looking for speakers and believe everyone has an interesting talk hiding inside themselves. If you are interested in letting that talk out at a future event you can get in touch at the meetup, on the website (https://melb.awsug.org.au/speak/), or reach out to one of the organisers on our Slack.