An LLM-Driven Multi-Agent Framework for Telescope Proposal Peer Review (arxiv.org)

A groundbreaking framework named AstroReview has been introduced to streamline the peer review process for telescope proposals, addressing the growing demand for observatory time in the field of astrophysics. This open-source, LLM-driven multi-agent system automates the review in three key stages: assessing scientific merit, evaluating feasibility, and performing meta-review for reliability verification. Notably, AstroReview achieved an impressive 87% accuracy in identifying genuinely accepted proposals during experiments, all without requiring domain-specific fine-tuning. The significance of AstroReview lies in its potential to enhance the efficiency and transparency of the proposal review process, critical to advancing astronomical research. By implementing iterative feedback through its Proposal Authoring Agent, the acceptance rate for revised drafts surged by 66% after two iterations, demonstrating clear improvements in proposal quality. This innovative approach paves the way for a more scalable, auditable, and high-throughput review mechanism, which could be transformative for resource-limited observatories struggling with an overwhelming number of submissions.