Join a Claude Community Event, organized by local enthusiasts for everyone who loves building with Claude!
A Claude community meetup for developers.
Registrations close at 1pm Monday 16 as we have to submit the registered names to AWS Security. Only approved attendees will be admitted.
Claude Code Community Australia is excited to host a Claude Code Meetup.
Presented by Dominik Fretz from Harbour Edge Intelligence.
Limited Spots. Good people. Local Builders.
Agenda
17:30 – 18:15 | Doors open, food, drinks and networking
18:15 – 18:25 | Welcome and intro
18:25 – 18:45 | From Vibe Coder to Builder: Shipping an AI App with Claude Code in 5 Days — Lingyun Zhao
Vibe coding gets you a demo. Getting to something you can trust takes more.
I'm building DrainWise solo with Claude Code for the City of Parramatta hackathon. It's an AI tool that predicts which stormwater drains will block next, so crews can clean the riskiest ones first.
In this talk, I'll show how I actually build with it:
• Setup and workflow: project brief, context handoff, keeping the agent on track
• Agents across the stack: frontend, backend, auth and database
• What broke, and what I'd do differently
• Next step: RAG and an AI system that learns from feedback
Takeaway: AI makes building fast. The workflow around it makes it shippable.
I'm a Sydney-based software engineer and indie founder with a PhD in Computer Vision from Queen Mary University of London and around seven years of full-stack experience. I'm building EvidMatch, an AI-powered talent marketplace, and DrainWise, a hackathon project that predicts blocked stormwater drains, both with Claude Code. I'm interested in AI-assisted development, practical AI adoption, and turning technical ideas into real products.
18:45 – 19:05 | The Benchmark Was Wrong: What Claude Found When We Gave It Context Instead of SQL — Pete Chapman
Enterprise data is the hill that AI agents still die on. Using BEAVER, MIT’s university data warehouse benchmark, agents writing SQL answered just 26% of questions correctly. Two-thirds of their answers were confidently wrong.
We ran the same questions through Claude Opus 5.5, GPT-6.1 Sol and Grok 4.7, this time connected over MCP to an Amorfs knowledge pack: a structured, typed projection of the same warehouse, with joins pre-resolved and business meaning made explicit. Accuracy rose to 97%.
Then something amazing happened. Asked simply to review the benchmark, Claude answered every question in three minutes and then flagged discrepancies in the official answers. That review led to 18 proposed corrections, now sent to the benchmark’s authors.
This talk covers how the pack is built, what agents need from context that schemas don’t give them, and the design patterns that bring context into your own MCP servers.
Pete Chapman is founder and CEO of First Cognition, builder of Amorfs, a context layer for AI agents. He has a background in cognitive science and enterprise architecture, and is based in Sydney and Ho Chi Minh City.
19:05 – 19:25 | One founder, an agent team: shipping a full business platform (web, iOS, Android, payments, door access) to production — Prashan Narayanasamy
Movena is an operating system for gyms, from a single location to multi-site groups: timetable and bookings across locations, memberships and Stripe payments, door access, a gym website with an embeddable booking widget, member apps on iOS and Android(HealthKit and HealthConnect), and a connector in the ChatGPT app directory and meta/google campaign connector for leads. It is built and run by one person.
This talk is about the operating model that makes that possible, not about prompts.
What I'll cover
-
Lanes, not one big chat. Separate Claude Code sessions own backend, review and infrastructure, each in its own git worktree, with rules about what each may touch, deploy or message.
-
The founder locks behaviour; agents don't. Every defect is walked through and its behaviour locked before any code is written. Codex implements from a paste-ready brief; Claude reviews.
-
Reviewers are not the final authority. How "individually reasonable" review suggestions quietly grew a feature from two levels to six, and the written rule that stopped it.
-
Two reviewers beat one. A real example from this week: my review passed a widget change; an independent review lane proved with a test that the new route was missing from a proxy allowlist and every Book tap would fail. It never reached a customer.
-
Guardrails that actually hold: deploys only on an explicit go in chat, contracts for released mobile APIs, one defect tracker of record, help pages shipped with every feature, and money changes tested end to end against the Stripe sandbox.
What you'll walk away with: a repeatable loop (walkthrough → lock → brief → independent review → UAT → production) and the specific rules I wish I had written down on day one.
https://movena.com.au
Prashan is the founder of Movena, an operating system for gyms from a single location to multi-site groups, spanning web, iOS and Android apps, payments and door access. He builds and runs it solo with a team of AI coding agents and ships to production several times a week.
19:25 – 19:45 | Stop benchmarking models, benchmark your harness — Jack Rudenko
Everyone benchmarks models. Almost nobody benchmarks the harness around the model, and that is where most of the variance in real work lives.
We built an internal eval rig to test our own Claude Code setup against stock Claude Code and other harnesses on real repos. Most of the config folklore did not survive contact with the numbers.
What I'll cover:
-
How to build a repeatable eval for a coding harness: task sets from real repos, graders, cost and wall clock as first class metrics, repeat runs so one lucky pass does not fool you
-
What actually moves the needle (context discipline, tool surface, loop shape) vs what does nothing or makes things worse (piling on MCP servers, a giant CLAUDE.md, subagents for everything)
-
Model vs harness: a frontier model bare against a cheaper model in a good harness, and what the price gap actually buys you
-
Live run: stock setup vs tuned setup, read the numbers on stage
Walk away with: a method to bench your own setup this week, the metrics worth tracking, and a shortlist of config changes that measurably pay off.
Format: 20-25 min with the live demo, or a 10-min lightning cut without it.
Co-founder and Chief AI Officer at MadAppGang in Sydney. Twenty five years of custom software delivery, now mostly building agentic harnesses, evals and tooling for teams running AI in production.
19:45 – 20:00 | Speaker panel Q&A
All four speakers take questions from the audience.
20:00 – 20:10 | Community announcements and close
What is Claude Code?
Claude Code is Anthropic's agentic coding tool that lives in your terminal. It lets you delegate coding tasks directly to Claude—whether you're building new features, debugging tricky issues, refactoring legacy code, or exploring unfamiliar codebases. Just describe what you want in natural language, and Claude Code handles the rest: reading files, making edits, running commands, and iterating until the job is done.
Think of it as pair programming with an AI that can actually touch your codebase. No copy-pasting between a chat window and your editor—Claude Code works where you work.
PHOTOGRAPHY DISCLAIMER: We will have a photographer on site to take photos for our community event page, and social media coverage.e
ID Required
You will need to have signed up and been approved to the meetup and bring ID (electronic is file) with you. This is a requirement of the building security.