Ciccarelli K1, Van Den Berg M1, Connor S1, Harris J1, Doherty S1, Vargas Ruiz C1, Gething P1,2
1The Kids Research Institute Australia, , Australia, 2Curtin University, , Australia
Biography:
Kai is a Software Developer at The Kids Research Institute Australia. They have over five years of experience as a developer working previously in the mining simulation industry and now in the research software development sphere. Her work at the Kids largely centres around working with geospatial data, developing cloud native applications, and creating AI solutions for research purposes.
Abstract:
Navigating the malaria research landscape involves synthesising fragmented data across surveys, reports, and geospatial repositories to answer complex and open-ended public health questions. Traditionally, addressing these challenges required time-consuming manual collaboration between epidemiologists, data scientists, and programmers. While researchers increasingly look to automate workflows, general-purpose LLMs and standard RAG systems lack the domain specificity and collaborative capacity required to resolve complex, multi-layered research questions.
This presentation showcases our progress towards the development of a multi-agent network framework designed to surpass the limitations of basic AI tools. Rather than relying on a single, generalised model, our approach aims to enable multiple specialised agents that actively collaborate to develop spatiotemporal solutions for malaria applications. The effectiveness of our approach is demonstrated across a diverse range of tasks, such as:
– Writing comprehensive literature reviews based on relevant papers with citations that link claims back to the specific text in the source.
– Extracting and harmonising complex variables across disparate population datasets into structured relevant formats.
By enabling researchers to guide these collaborative agent networks, the framework reduces reliance on intermediary programming teams and accelerates complex workflows. Furthermore, dividing tasks among specialised agents significantly reduces hallucinations, while visually traceable reasoning logs allow researchers to audit the collaborative logic of the network.
The presentation examines the orchestration architecture of these agent networks, alongside development approaches, security considerations, and implementation challenges. By evaluating researcher perceptions of time savings and current barriers to adoption, we outline a scalable blueprint for applying multi-agent spatiotemporal solutions across data-intensive research domains.
