AI assistants are being developed for research and industry bodies, and businesses, to access tens of thousands of pages of often buried information in one place.
Among them are the on-farm answer dog, Bella, for Beef + Lamb New Zealand (BLNZ), CultureQ for indigenous technical company Kiwa Digital and AI assistant Ask Farai for the Foundation for Arable Research (FAR).
Data largely going untapped is now being made more accessible by the generative AI platform.
More pilots are being built for clients in Australasia and potentially the United States.
Custom D was co-founded as a spin-off from Digital Fusion, owned by Craig Saunders, who joined chief executive Julie Ryan, Sam Sehnert and Matthew Rhodes as original shareholders in 2014.
Another three shareholders were added with further investment to be raised this year.
Mrs Ryan said the AI platform would be scaled up this year after building over the past 18 months.
Demand was highest among research groups wanting to get research to their users.
"We were in the US [in December] and in the US alone there is $1 trillion [$NZ1.7b] worth of research conducted each year. ... A study says only 14% of that research is being used and this is research that can save lives, reduce human impact on the planet and improve our food production."
AI technology was moving at breakneck speed, with many people now using ChatGPT, Mrs Ryan said.
But building a system maintaining integrity at scale involving tens of thousands of documents with "guard rails" was more difficult.
Custom D saw the potential for their customers when large language models such as ChatGPT, Gemini, Claude and others emerged, she said.
It started working with FAR in 2024 to deliver 30 years worth of research data to growers so they could lift crop yields and improve farms.
"That could be assisted by this research, but the research is technical, it’s quite long and hard to get to as these documents are multi-page PDFs with tables and lots of text in scientific terms," Mrs Ryan said.
Generative language models combined with retrieval-augmented generation absorbed and accurately summarised vast amounts of data so it made sense to people.
The Ask Farai assistant went into production later that year, designed to make it secure from hackers.
Questions fielded to Caitlyn — once known as "AI Chatbox thingy" — are returned with answers and a link showing the information source.
"The important part for Caitlyn is you control the knowledge source. So where you have the likes of Gemini or ChatGPT and all of these other ones, you are sending your knowledge to these models outside of your organisation, whereas we deploy this into our customers’ own private cloud environment."
Customers can delete or add data, instead of relying on the web for answers.
Interest generated by the FAR assistant increased when BLNZ launched its tool last September, Mrs Ryan said.
"That sent another flurry of activity in that agricultural research space. We have a number of research clients now in NZ and we are talking to one of the largest CROs in Australia and are about to start a pilot with them and have nearly completed a pilot for another one of the very large research institutions here in NZ."
Encouraging talks were had with agricultural leaders during the US visit and a consolidation of research institutions in North Carolina and Boston showed promise, she said.
Caitlyn has a multi-lingual function for non-English speaking countries.
Last year, Custom D won innovation partner of the year and social impact partner of the year at awards during the AWS re:Invent conference in Las Vegas.
A 25-strong team includes developers, cloud architects, data scientists, project managers and marketing and sales staff.
The company has about 10 Caitlyn projects at the pilot stage and another 10 in production for the likes of Deer Industry NZ and NZ Avocado and Forest Growers Research, as well as Birchip Cropping Group in Australia and Australian crop and seed distributors.
Mrs Ryan said businesses such as input providers and manufacturers carrying out lots of research would benefit from Caitlyn, as well as public and government agencies.
More than 15,000 engineering hours went into building the AI platform.
Future functions could include connecting to other parts of a system or tools to build more work flows, she said.
"What can be done on top of that — the world is your oyster."
