UC Riverside Undergraduate Research Journal

The growing volume of scientific literature presents a significant challenge for researchers to identify relevant methods, compare findings, and build upon existing work. Although advances in artificial intelligence have led to new tools for literature analysis, many of these tools rely on generative models whose outputs are not always grounded in verifiable sources. This raises concerns about reliability and reproducibility in scientific research. In this study, we explore whether a retrieval-based system can provide more transparent and trustworthy support throughout the research process.

We present Chem-AI, an interactive tool that structures the research process into clear steps, including query interpretation, literature retrieval, experiment classification, and method mapping. Unlike conventional AI systems that generate answers without clear justification, Chem-AI links each output directly to published research, allowing users to trace and verify the information.

The system was evaluated using a diverse set of chemistry research queries spanning reaction mechanisms and analytical techniques. Compared to standard generative models, Chem-AI retrieved more relevant sources (with roughly 95% of responses rated as fully relevant), produced responses grounded in established methods, and reduced unsupported or unverifiable claims. These results suggest that a retrieval-based approach improves both the reliability and clarity of AI-assisted research.

By aligning with the structure of real research workflows, Chem-AI provides a more transparent and user-centered way to navigate scientific literature. This approach is especially useful for undergraduate researchers, supporting more accurate, reproducible, and methodologically grounded work.