AI Engineer

Title: AI Engineer | Build the systems that run a 100-person business | Melbourne hybrid

Introduction

Most AI roles are a proof of concept that dies in a slide deck. This one is not.

Learning Online Group is an online education company operating across Australia, Canada, the UK and New Zealand. Over 100 staff, a portfolio of brands spanning beauty, photography, animal care and tattooing, and a growing amount of the business quietly running on systems our AI function built.

That is not aspirational. We already have LLM-powered document automation, AI-assisted grading and chatbots in production around Salesforce, n8n and Postgres. What comes next is planned and funded: a Google Cloud data platform on BigQuery, and a multi-agent orchestration layer powering tutoring, student service and analytics agents across every brand.

Somebody has to build that. We are hoping it is you.

Duties & Responsibilities

You are an engineer first. Not research, not data science. You write tested, typed Python, ship behind CI/CD, and own what you run in production. When it breaks at 7am you are the one reading the logs.

You will work across Google, Salesforce, Vonage, Customer.io and WordPress, connected through n8n and custom apps, and increasingly through AI agents and tools like Claude Code and Antigravity.

Depending on where you land in the range, the job looks like this:

• Build and operate production LLM applications: RAG over governed knowledge bases, tool and function calling, structured outputs, and multi-agent orchestration where a supervisor agent coordinates specialists
• Own evaluation and observability: golden datasets, automated evals for non-deterministic behaviour, tracing, cost and latency monitoring. If you cannot prove it got better, it did not
• Build integrations and semantic data layers across Salesforce, our LMS and finance platforms, with event-driven automation and APIs that other people can actually use
• Stand up our Google Cloud AI stack: BigQuery, Cloud Run services in Python (FastAPI), vector search, and the Claude, OpenAI and Gemini APIs
• Implement guardrails and responsible AI controls: least-privilege access, prompt-injection defence, privacy-safe handling under the Australian Privacy Act
• Monitor production, triage failures, and keep the runbooks honest

The first six months are already mapped: take ownership of the live automation estate, ship internally scoped features, take our internal knowledge assistant from design to production in Slack, and lay the BigQuery foundations for the agent roadmap.

Desired Experience & Qualification

We have deliberately advertised one role with a wide band, because we would rather meet good people than argue about titles. Where you land depends on what you have actually shipped.

To be in the conversation at all:

• Solid Python, with Git, and enough SQL to check your own results
• You have called REST APIs and handled JSON and webhooks in something you built
• You have built at least one thing with an LLM API and can explain, in detail, where it breaks. Personal projects count. Hackathons count. "I read about RAG" does not
• A degree in computer science, software engineering, AI or data, or the equivalent proven in work

To be at the top of the band:

• 5+ years software engineering, including 2+ years running LLM or agentic systems in production, where "production" means real users and real consequences
• Hands-on RAG: embeddings, vector databases, retrieval design, and the evals that prove quality
• Enterprise integration depth: REST, webhooks, OAuth or JWT. Salesforce experience is a real advantage
• Cloud engineering with Docker and CI/CD. Google Cloud preferred, AWS or Azure welcome
• You have set engineering standards for other people and are happy mentoring a less experienced engineer
• You can sit with a non-technical department head, work out what they actually need, and translate it into a system

Nice to have at either end: multi-agent frameworks (LangGraph, CrewAI), Model Context Protocol, FastAPI, education sector experience.

Some of our strongest applicants come from product engineering at Australian scale-ups (Canva, Culture Amp, SafetyCulture, Employment Hero, Airwallex), consultancies and delivery shops (Thoughtworks, Mantel Group, Deloitte Digital), Salesforce partner work, or from being the person who quietly automated half an operations team's job somewhere nobody asked them to.

What matters most:

• Self-directed. You do not need a detailed brief to start moving
• Motivated by impact. You want to watch the thing you built change how a business operates
• Comfortable with ambiguity. This function is still being defined and you get a say in it
• Commercially minded. You ask whether the build is worth the time before you start it
• Fast learner. The tooling moves weekly and you enjoy that rather than resenting it

Package & Remuneration

$105,000 to $165,000 AUD, and we mean the whole range.

That is deliberately wide, so here is exactly what it means:

• Around $105,000 to $120,000 is for someone early in their career who has genuinely built with LLM APIs, can work through an integration without hand-holding, and wants to grow fast. You will work alongside senior engineers, ship with review, and take on more as you prove you can
• Around $130,000 to $165,000 is for someone who has run production AI systems before, can own the architecture, set the standards, and mentor someone more junior. You take the function and you run it

If you are somewhere in the middle, apply. We will work it out together.

• Full-time, permanent
• Hybrid working from Melbourne
• Reports to the CTO

Why people love this role

• Real scope and hands-on from day one, with real budget and real stakeholders
• Direct access to the CTO and leadership team, not five layers of management
• Your work ships in weeks, not quarters, and people notice
• Exposure to a $6M+ paid media operation across four markets

Why people stay

• We invest in learning. Tools, courses, conferences, if it makes you better at your job we want to hear about it
• We back people early. Several of our senior team joined young and grew into it, and we are actively looking to do that again
• You get to build production AI properly, with evals and guardrails, instead of shipping demos
• The problems are real operational problems with measurable outcomes