Mephisto: Self-improving Large Language Model–based Agents for Automated Interpretation of Multiband Galaxy Observations

Abstract Astronomical research has long relied on human expertise to interpret complex data and formulate scientific hypotheses. In this study, we introduce Mephisto—a multiagent collaboration framework powered by large language models (LLMs) that emulates human-like reasoning to analyze multiband galaxy observations. Mephisto interfaces with the CIGALE codebase (a library of spectral energy distribution models) to iteratively refine physical models against observational data. It conducts deliberate reasoning via a tree search, accumulates knowledge through self-play, dynamically updates its knowledge base, and is validated across diverse galaxy populations—including the James Webb Space Telescope’s recently discovered “little red dot” galaxies. We show that Mephisto demonstrates the ability to infer the physical properties of galaxies from multiband photometry, positioning it as a research copilot for astronomers. Unlike prior black-box machine learning approaches in astronomy, Mephisto offers a transparent, human-aligned reasoning process that integrates with existing research practices. This work underscores the possibility of LLM-driven agent-based research for astronomy, establishes a foundation for fully automated, end-to-end artificial intelligence (AI)-powered scientific workflows, and opens up new avenues for AI-augmented discoveries in astronomy.