A new paper from Unilink’s AI Research Centre is now informing work around inmate activities, interventions, and the wider inmate journey.
Unilink’s AI Research Centre has produced a new paper, MAGIC: Multi-Agent Generative Intention Coordination, which was presented at the top-ranking (CORE ranking A*) International Conference on Learning Representations (ICLR 2026) by David Huk, who led the work during his internship.
Co-authored by Oliver Hamelijnck, Dimitris Demiris, and Theo Damoulas, the paper explores coordinated behaviour among agents and how to model it at scale. At its core, the paper highlights the problem many in the justice system face in capturing and coordinating the activities of multiple agents.
Across product development at Unilink, the paper supports our work in exploring how the code it covers can be used in Unilink products, particularly around inmate activities.
Our aim is to help the system learn more about which activities may be suited to individual inmates. In practice, this could mean giving the system a stronger basis for recognising patterns across behaviour, activity, and intervention history, rather than treating each point in isolation.
This also connects to our work around reoffending prediction in the context of interventions. We are looking at how the same line of research could support a more informed view of how interventions relate to outcomes over time, and how that can sit inside a fuller picture of the inmate journey.
In practical product terms, activity management, intervention planning, and prediction are not siloed in day-to-day use; if the system is truly going to be useful, the setup should reflect that.
This is the type of work for which the AI Research Centre was set up, with a key focus on research that can be used inside real products and applied in live justice environments. That means technical work has to stand up academically, but it also has to be relevant to operational systems and the way they are actually used.
Now, the next step is development and testing. We are now exploring how this research can support product thinking, how it performs in practice, and what it can support across inmate activities, interventions, and the wider journey through the system.
Read more about Unilink’s AI Research Centre here.