About Novastra
Built by an engineer who'd rather
show you the system running
than the slide deck describing it.
Riley Ball started Novastra because he kept seeing the same pattern: businesses would engage an AI consultancy, get a beautiful slide deck, and then nothing would actually make it to production.
The problem is usually the same. The people who understand the business don't understand the engineering, and the people who understand the engineering treat every project like an academic exercise. The result is a lot of promising demos that never survive contact with real data, real users, and real infrastructure constraints.
His background is in building things that have to work. Aris Detect -- the SaaS platform he built and runs -- processes drone-based property inspections for real inspectors who depend on it daily. It includes ML-powered damage detection, GPU inference pipelines, real-time collaboration, and autonomous AI agents that handle bug triage and code review 24/7.
“If your problem doesn't need AI, I'll tell you. If it does, we'll find the simplest approach that actually works.”
-- Riley Ball, Principal Engineer
That experience shapes how he approaches consulting. Start with the problem, not the technology. Build for production from day one -- not “we'll worry about deployment later.” And stay involved through deployment and beyond, because shipping is where the real value is.
Riley Ball
Principal AI/ML Engineer
Brisbane, Queensland
The best AI projects are the ones where someone says “I didn't know that was possible” — and then uses it every day.
How we think about AI consulting
Start with the problem, not the model.
If your problem doesn't need AI, I'll tell you. If it does, we'll find the simplest approach that actually works.
Build for production from day one.
No throwaway prototypes. Everything we build is designed to deploy, scale, and maintain.
Practical over impressive.
A system that saves your team 10 hours a week is worth more than a cutting-edge model that's too fragile to run unsupervised.
Stay until it works.
I don't hand off a repo and disappear. We work together until the system is running in production and your team knows how to maintain it.
Technical depth
We go deep on a few things rather than wide on everything.
AI consulting is full of generalists who can set up a pre-trained model but can't debug why it fails on your data. We prefer to work in areas where we have genuine production experience.
Computer vision and object detection
YOLO, MMDetection, PyTorch. We've trained, evaluated, and deployed custom detection models for real-world use cases -- not just run pre-trained weights through a tutorial.
GPU infrastructure and model serving
RunPod, AWS Lambda, Docker. We know the difference between GPU provisioning that works for a demo and GPU provisioning that works at 3am when your pipeline spikes.
Full-stack application development
React, Next.js, Laravel, TypeScript, WebSockets, PostgreSQL. ML models are useless without the application layer that puts them in front of users.
Agent orchestration and LLM integration
Claude, GPT-4, MCP, Discord bots, webhook pipelines. We build and operate autonomous agents daily -- not as experiments, but as production infrastructure.
Based in Australia
Same timezone. Same context.
Working with someone who understands Australian business culture, regulatory context, and time zones makes everything simpler. No midnight standups, no lost-in-translation requirements.
Ready to talk about your project?
It starts with a conversation. Tell me what you're working on, and I'll tell you straight whether I can help.
Start a conversation