Digital Futures Institute: Case Studies

Diometer

Farms of every kind have significant administrative overheads, whether this is adhering to approximately 150 different pieces of legislation, applying for sustainability related funding or completing audits. Farms spend on average two person-days completing each report for supermarket/certification audits alone.

Diometer, an East of England SME agritech business, seeks to address this problem through its modular Flexifarm software platform and helping farms to digitise costly manual and paper-based processes which currently underpin the approach of most farms to handling compliance needs.

The University of Suffolk team of Professor Nicholas Caldwell and Dr Salman Ahmed have been working with Diometer since October 2024 on two related projects to identify and develop Artificial Intelligence solutions to enhance Diometer’s Flexifarm platform.  As the data and compliance requirements of farms vary according to crops, customers and the farms themselves and over time, Diometer has needed to adopt databases based on the NoSQL paradigm rather than the relational paradigm (e.g., SQL, MySQL, Access, etc). NoSQL is better able to cope with variety, flexibility and data change but more difficult for non-experts to write database queries. The University team have leveraged large language models to create a prototype which converts English language queries into accurate NoSQL queries and then summarised the database results back into natural language. Additionally, the team have been able to automatically generate completed forms from natural language queries.

An unexpected outcome of this work was proving that Large Language Models (LLMs) do not always improve with more recent versions. For working with NoSQL, the team found that ChatGPT 3.5 Turbo had optimal performance in terms of number of tokens used, price per token, time to respond (average time approximately 0.8 seconds) and 95% accuracy on test queries. ChatGPT 5 versions used an order of magnitude more tokens and cost, had response times ranging from six to 31 seconds, and demonstrated 60-70% accuracy on the same test queries. This research has concluded that GPT-5 models are more ‘rule-literal’ than their predecessors, and this rigidity undermines reliability in practical settings where schema variability and natural phrasing are common.

The University team are continuing to collaborate with Diometer to improve this automated form-filling capability, and to develop additional AI solutions for other farming problems.

This work has been supported by Innovate UK grants: Agri-F2P: Feasibility Project for AI-Driven Forms Management in Agriculture (number: 10122783) and  Agri-KG: Transforming Agriculture with an AI-Powered Knowledge Graph (number: 10129638).

Discover More

The University of Suffolk enjoys a long-standing collaboration with the charity, Suffolk Together, and their spinout company, Discover More (Suffolk) Ltd, previously known as Suffolk Libraries Industrial and Provident Society and responsible for library provision across  Suffolk.

The underlying assumption of the original work was that library usage and participation in library-hosted activities can have a positive outcome on mental health, wellbeing, happiness and reducing social isolation and loneliness. The technical hypothesis was that an AI solution could be developed to support the matching of activities to needs.

In 2019– 2020, Professor Nicholas Caldwell supported Suffolk Libraries in winning a NESTA Tech to Connect award, which Suffolk Libraries used to fund the University of Suffolk to: a) run focus groups to establish whether end users would provide information on wellbeing needs (because if they were unwilling to do so, no solution would be possible); and b) undertake usability tests on an initial design mock-up.

In 2021–22, Professor Caldwell, Dr Abbie Millett and Professor Sarah Richards won a KEEP+ funding award to work with Suffolk Libraries to: a) investigate if there was statistically significant evidence to corroborate positive outcomes on wellbeing and loneliness through activity engagement; and b) to create an AI-based recommendation engine using the most appropriate combination of techniques to match wellbeing needs with suitable activities. No significant findings were identified for loneliness, however, as activity group use increased, social connectedness and social assurance also significantly increased. There was no significant effect on happiness, however, there was a significant increase in satisfaction with life and mental wellbeing after attending an activity group. Collectively these results demonstrated that Suffolk Libraries has a significant positive effect upon its users in terms of both eudaimonic and hedonic wellbeing. The computing elements of the project successfully delivered a prototype encompassing Natural Language Processing and mapping algorithms for labelling events organised by Suffolk Libraries, as well as a recommendation engine linking users to events based on each user’s individual emotional needs. The system was specifically designed to support recommendation of activities from a dynamic set of activities rather than a tightly curated set of services. The two halves of the project provided Suffolk Libraries with a working prototype and evidence base enabling them to secure their Arts Council funding to develop a deployable version of the software, known as Discover More, which has since been deployed live.

In the first five months of the Discover More service being deployed on the Suffolk Libraries website in 2025, there were 2,156 account holders, 25,000 events and activities were showcased and the service delivered 20% improvement in wellbeing scores.

With the transfer of the county libraries contract away from Suffolk Libraries, they have since evolved into the Suffolk Together charity and the spinout company Discover More (Suffolk) Ltd, which is taking forward the commercialisation of the Discover More software.

This has since evolved into the Discover More online service (run by the spinout company). Our current work is focused on creating data curation agents to provide the hyper-localised activities that the service recommends, and supporting the company and the charity in positioning the Discover More service for commercial applications such as social prescribing in the NHS, student support in the university sector and workplace wellbeing.