Human-Centred Research Automation: Agentic AI Framework for Literature Discovery, Filtering, and Gap Identification

Recent advances in agentic AI raise opportunities to automate complex research tasks; however, the EU AI Act and Human-Centred AI (HCAI) principles emphasise that such automation must safeguard trust, transparency, and responsibility. This work-in-progress study developed an automated framework using agentic AI to support SLRs, which are critical for synthesising knowledge and identifying research gaps but remain resource-intensive, time-consuming, and in need of trust and transparency in their processes. The study presents a human-centred, agentic AI framework for discovering, filtering, analysing academic papers, and identifying research gaps. The proposed framework combines machine efficiency with human oversight to ensure contextual interpretation and responsible judgment. In a pilot study using Llama 3.1 8b, it achieved 71.05% accuracy for abstract filtering compared to a human-led SLR baseline. It also generates structured outputs, such as research gaps, potential directions, and candidate solutions, that enhance clarity and evaluability of the automated synthesis. Preliminary findings demonstrate that automation makes the initial synthesis faster but the human involvement in the later stage ensures that the result can be reliable and transparent. To further assess the framework’s alignment with HCAI principles, the next phase will conduct qualitative researcher interviews to evaluate perception of trust and transparency, as well as measuring how structured outputs improve interpretability. Overall, the study contributes to the design of responsible and trustworthy AI systems for research, demonstrating how agentic AI can make SLRs more efficient, reliable, and accessible, while advancing discussions at the intersection of agentic AI, systematic reviews, and human-centred research practice.