When building it yourself is the only option

In the wake of the Mycoplasma bovis outbreak in cattle herds, and canker blighting kiwifruit vines, our primary sector is on high alert for potentially devastating biosecurity threats.

If the big one was to hit – a large outbreak of foot and mouth disease requiring widespread culling of dairy and beef cattle herds – the impact on the New Zealand economy could be $10–$15 billion.

Veterinary epidemiologist Dr Tom Brownlie was a young vet in 2001, working with cattle in the dark days of the UK’s devastating foot and mouth outbreak. More recently, he helped the Ministry for Primary Industries during the Mycoplasma bovis outbreak, looking at data from so-called index farms to try to piece together how the disease had spread.

“Using some rudimentary statistical tools, not any artificial intelligence, I could start to see patterns in those agricultural data, about 18 months earlier,” says Brownlie, an expert in his field who advises the UN’s Food and Agriculture Organisation.

“Mycoplasma bovis was probably in the country for two years before that vet in Oamaru diagnosed it.”

First identified in 2017, the presence of the bacterial disease on New Zealand farms required the culling of nearly 184,000 cows and cost $800 million. The economic impact, not to mention the emotional toll on farmers, could have been substantially reduced if we’d discovered it sooner.

An early warning system for animal disease

The event spurred Brownlie and Dr Christopher Laing, chief technology officer at AI and data analytics specialist Qrious, to develop an AI-based early warning system for animal disease.

Sentinel-AI emerged a year ago as a real-time monitoring tool powered by machine learning, and drawing on a wide range of data sources from government agencies, including the National Animal Identification and Tracing (NAIT) programme, and agritech companies testing factors such as milk quality.

Sentinel-AI has been well received by the agricultural sector, and the model underpinning it is a world first. But Brownlie, who founded Ingenum in 2016 as a data-driven animal epidemiology startup, encountered a major problem during the course of 2022.

He and Laing realised that there were too many gaps in the data sources available to Ingenum, to build the early warning system the country needed.

“We peeked under the hood at some of these agritech companies and it turned out they didn’t have quite what they thought they did,” says Brownlie. “We realised that there were bigger holes in the data sets than we expected.”

Sentinel-AI was missing the most important input of all – the clinical observations farm vets were making every day, all over the country.

“It might be along the lines of, ‘I saw a beef heifer with acute respiratory signs and some ulceration in the mouth and I gave it antibiotics and some painkillers. I'll see it again next week’,” says Brownlie.

That’s exactly the sort of information that a ChatGPT-type service, underpinned by a large language model, could draw on to flag emerging signs of infectious disease. It led to a major “pivot in focus” for Ingenum. Veterinary practice software providers weren’t moving quickly to take a more data-driven approach to collating information on farm animals.

“I know from my own experience as a vet that my clinical notes would go into my diary or a carbon copy pad and sit in my truck for a few days until I had time to enter them into a computer,” says Brownlie.

VetGPT

The missing link, it turned out, was an app to make collating clinical notes easier for vets, in the process generating the motherlode of geographically distributed data that would allow Sentinel-AI to spot early warning signs of disease.

The New Zealand Veterinary Association, which has over 3,400 members nationwide, came onboard as an enthusiastic supporter. The machine-led veterinary information system  (MaVIS) developed with the help of Ingenum’s technology partner Qrious, is now being piloted by vets. It also serves as an invoicing tool, allowing AI-generated clinical notes and bills to be written up in the same interface.

As Laing explains, Ingenum had joined a long list of companies needing to develop a new product, or even a whole technology platform, to realise their primary goal. Ecommerce giant Amazon built its own hosting infrastructure in the early 2000s to support its rapidly growing online store, and in the process spawned AWS, now the biggest public cloud platform in the world.

“It was similar with the creation of the Go programming language,” says Laing.

“Google had created a problem for itself no one had before: how do you serve billions of requests per second on a website? No one had a website that popular before.”

 Netflix was in a similar boat, needing to develop a new compression algorithm to accommodate the scale of its rapidly growing streaming platform.

“If we waited until someone else solved this problem for us, someone else would probably solve the problem instead of us but it would take a very long time,” Brownlie concludes.

“We had to be the ones to step up.”

MaVIS also indexes tens of thousands of peer-reviewed scientific articles, so will present clinical information for vets in response to text or verbal prompts, which can be added to case notes and presented to testing laboratories as well.

“Now we can have everything from weather data, soil data, data coming in from agritech sensors, to NAIT data and the clinical science coming in from vets,” says Brownlie.

“If we can train the model to continuously learn to detect explicit disease signs, then that translates quite easily across different types of agriculture and horticulture,” he adds.

Ingenum now has a business model that derives revenue from SaaS subscriptions, which also gives Sentinel-AI a more secure financial footing. Ingenum has been supported by Callaghan Innovation grant funding to date and is about to undertake a seed capital raising round.

“The next 12–18 months is about fine-tuning Sentinel-AI and launching MaVIS as a product that makes life easier for vets, while aggregating data that could form the basis of our early warning system for disease,” says Brownlie.