Agentic Workflows for Gap-Aware Literature Reviews

Effective literature reviews are foundational to scientific discovery. However, traditional approaches often struggle with coverage, bias, and difficulty in surfacing under explored areas. As AI systems become increasingly integrated into research workflows, there is growing potential to make them proactive collaborators than passive research assistants. In this paper, we introduce an agentic workflow for conducting gap-aware literature reviews that combines structured synthesis, knowledge graph modelling, and perspective-guided questioning to enhance both comprehensiveness and traceability.

Our approach leverages a multi-agent pipeline that generates outline-driven literature reviews informed by user-specified goals or perspectives, while concurrently identifying gaps in reasoning, coverage, or evidence using graph traversal and contrastive retrieval techniques. To support graph traversal, we first construct topic-specific knowledge graphs from scientific corpora, enabling structured reasoning over the current state of knowledge. The system integrates citation fidelity and source attribution mechanisms throughout, helping ensure transparency and reproducibility.

We demonstrate the effectiveness of this workflow in scientific domains where contextual awareness and information completeness are critical. The initial results provide evidence that shows our approach leads to more focused, structured reviews and enables researchers to identify neglected but important areas for further investigation proactively. By operationalising literature analysis into an interactive, explainable agentic workflow, this work lays the foundation for AI-led collaborative early-stage scientific research.