iscover Generic AI wasn't good enough. So, they built their own.
TechnologyElfworks co-founders Jimmy McPhedran and Ian Youngman built the platform after realising generic AI tools couldn't handle the complexity of Australian tax, here's their story.
When Jimmy McPhedran and Ian Youngman sat down to build an AI platform for Australian accountants, they were solving a problem they'd both hit from opposite directions.
McPhedran had spent 25 years in the profession, across big four firms, mid-tiers, and running his own practice. He'd watched Australian tax law become harder to navigate every year, with clients growing more complex and a regulator that shows no signs of easing up. "The job of an Australian accountant is becoming harder," he says, "and I see AI as really the only thing that's going to help us through the next few years."
Youngman came at it from the other side. With 20 years in digital transformation across global roles, he'd seen the rise of generic AI tools like ChatGPT and immediately understood their potential. But he also saw the risk. "There's a huge amount of risk and danger when dealing with complex subject matters that require very specific information," he says. Generic models, trained on everything and nothing in particular, simply weren't built for the density of Australian tax law.
That gap, between what generic AI could do and what the profession actually needed, became the starting point for Elfworks.
The pair set out to build something different: a platform that took the power of LLM’s and pointed them directly at the material that matters, legislation, case law, and ATO public releases. Rather than relying on a single model's judgement, they built the platform around multiple models working together, each with its own strengths and blind spots. "Every model out there is biased," McPhedran says. "If you get multiple biased models and get them to bash up against each other, it helps to eliminate the individual biases."
That approach is where the name comes from. After testing multiple models together and seeing the results improve, the founders landed on the idea of a group of helpers working behind the scenes, doing the heavy lifting an accountant doesn't have time for. "The magical helpers doing the grunt work for you behind the scenes," is how McPhedran describes it. Elfworks was born.
The first version, by McPhedran's own admission, wasn't much to look at. "If you're not embarrassed by your first release, you released too late," he says. What it lacked in polish, it made up for in accuracy, the priority from day one. Data security and precision came first, with the interface catching up later. Over the past six months, the platform has been substantially rebuilt and re-skinned, but the underlying focus hasn't shifted.
For Youngman, the technical challenge always came back to two things: eliminating hallucinations by grounding every answer in a solid, purpose-built database, and protecting client data through sovereign models hosted and controlled in Australia. "You need to make sure your data is not leaked into these models for training, or sent offshore for processing," he says.
That combination, deep functional knowledge of Australian tax paired with a genuine handle on the technology, is what the pair set out to build from the beginning.
See what Elfworks can do at elfworks.ai
Want to see more stories from trusted news sources?Make Accountants Daily a preferred news source on Google.