Broonie: local generative AI for schools

Thu, 01 Jan 2026 00:00:00 +0000

Schools keep a lot of data about their pupils and staff are required to regularly report on pupil progress and attainment. This task is repetitive, but time-consuming, because it requires integrating data in different formats from many sources.

AI tools could educators integrate and make sense of the data they have – but sending pupil data to a third-party AI service is a data protection problem most schools cannot responsibly take on.

Broonie’s approach was to run AI models locally instead. We aimed to help schools improve their learners’ attainment and reduce administrative workload without the data protection risk, by enabling them to use AI applications for data analysis on their own premises completely offline, without sending any data to third parties. The offering combined specialised hardware, software, and support.

I was one of three co-founders, in a team that grew to five, from September 2024 until the venture wound down in January 2026. My role covered teacher consultations, data analysis, ensuring compliance with AI, copyright, and data protection legislation, prototype implementation, user evaluation, and business administration.

We secured £26,100 in grant funding to develop the initial prototype from multiple competitions. We won the Young EDGE category of the competition and we took part in the Edinburgh Innovations Venture Builder Incubator 5.0, the Summer Start-up Accelerator, and the Converge KickStart Challenge, through which I took extensive training in entrepreneurship, marketing, pitching, user research, and working with commercial stakeholders.

Gig workers' personal data on the decentralised web

Fri, 01 Sep 2023 00:00:00 +0000

Over two research internships – August to November 2022, and July to September 2023 – I worked with the in the Department of Computer Science at the University of Oxford, on the Oxford Martin School’s (Ethical Web and Data Architectures in the Age of AI) project.

The question we worked on was how gig workers – Uber drivers, Amazon workers, Deliveroo riders – might take control of, and make better use of, their personal and work data in a setting shaped by platform power.

The first internship was a participatory design study: user interviews and co-design exercises about gig worker-led data sharing initiatives, with workers and the organisations that support them. I conducted some of the interviews, coded them, and analysed the findings to outline potential sociotechnical models for platforms.

The resulting paper examines how researchers might empower gig workers to steer how web decentralisation gets implemented in platform work, rather than having it done to them. It was published at CHI 2023: .

Decentralised generative AI

The second internship was technical: designing, building, and evaluating app prototypes using , the Social Linked Data protocol. Gig workers can request their data from platforms under GDPR subject access requests, but the responses arrive in incompatible schemas and unhelpful formats. The main prototype used large language models (small enough to run locally so the data never leaves the user’s control) to automatically integrate datasets with mismatched schemas, and to produce map and timeline visualisations from DSAR response data. Some of that work is open source at .

That line of work later led to SocialGenPod, a Solid-based architecture for privacy-friendly generative AI applications: user data – chat history, app configuration, personal documents – stays in the user’s own Pod rather than being tied to a model or application provider, with Solid’s access control deciding who can read it. It was published as a demo at The Web Conference 2024: , with the prototype at .

Personal Data |

Broonie: local generative AI for schools

Gig workers' personal data on the decentralised web

Worker-led data sharing

Decentralised generative AI