Aussie AI health-tech Heidi aims to cure clinical burnout

For a company that has saved doctors some 43 million hours of administrative work, Melbourne-born health-tech startup Heidi had a surprisingly bumpy start.

Six years ago, co-founders Yu Liu and Dr Thomas Kelly built a medical chatbot to train medical students on how to do consults with patients. When that failed to gain traction, they pivoted to a general practitioner (GP) platform offering telehealth, medical consultation bookings, and prescriptions.

“We learned a lot from that experience,” Heidi’s chief technology officer and co-founder Yu Liu told Computer Weekly in a recent interview, adding that the company initially struggled to pitch its software to clinics that were already using more established applications.

While the customers that adopted Heidi’s GP platform saw productivity gains of up to 20%, it wasn’t enough for Liu and his team. “My belief is that if you want to change something that is currently working, you need to be more than 100% better,” he said.

That realisation led to a second pivot towards transcribing medical consultations and automating clinical documentation – the core focus of the business today. Now valued at $465m and backed by $100m in total funding, Heidi’s cloud-based platform is used in over 110 countries, processing over two million consultations each week.

In February 2026, the company acquired UK-based clinical AI pioneer AutoMedica to expand its footprint in Europe. It also launched a suite of new tools: Heidi Comms for automated patient outreach, and Heidi Evidence, an ad-free clinical research tool built in partnership with medical databases and journals like BMJ to provide doctors with trusted, verifiable research at the point of care.

Heidi’s most ambitious move is a foray into hardware. The company recently launched Heidi Remote, a 21-gram wearable AI microphone designed specifically for the high-decibel, chaotic environments of hospitals.

“I believe everyone will have a remote AI wearable in the next three to five years,” Liu said. “Currently, when you are walking around, you will lose a good chunk of your clinical session because your mic is a conference mic or you have to hold your phone, which is super inconvenient. And doctors don’t always have good Wi-Fi.”

The device captures audio offline and syncs automatically, boasting a 14-hour battery life. But for Liu, the hardware is a stepping stone to a broader vision: eliminating the computer from the consultation room entirely.

“Some 30% of a doctor’s time is spent inputting what they gather in their head into computers,” Liu explained. “We want to remove computer use and combine a voice interface which converts prompts into tasks. You can tell it: ‘I’ve finished seeing this patient. Check this BMJ article for the dosage, and if it’s fine, open my prescription software and put the medicines in.’ One minute later, everything is there for the doctor to review and confirm.”

To power this level of complex automation, Heidi has changed how it builds its artificial intelligence (AI) capabilities. While the startup initially relied on off-the-shelf foundation models, high latency, rising costs, and a lack of clinical optimisation drove the company to build its own models.

Today, more than 80% of Heidi's workloads run on its own models. The startup uses a unique blind A/B testing system, presenting two AI-generated outputs side-by-side to doctors to objectively measure which model performs better in real-world clinical settings.

However, Heidi hasn’t abandoned frontier model providers entirely. It maintains a close partnership with Anthropic, using its Claude models for complex reasoning tasks performed by Heidi Evidence, prototyping new features and generating synthetic data to fine-tune the models.

As the company scales, with cumulative consultations growing 20% month-on-month in markets like France, Germany, and Singapore, Liu is eyeing an infrastructure play that moves AI away from the public cloud.

Recognising that data privacy remains the primary hurdle for mass hospital adoption, Heidi is developing an on-premise strategy that would see its entire AI software suite packaged into a physical server box and shipped directly to medical facilities.

“No one wants to send their data somewhere else,” Liu said. “A hospital can start using Heidi, and the data never leaves their site. We believe we have at least a six-to-12-month advantage over the competition in delivering this.”