AlphaAgent skill-contracted AI beats traditional RAG in materials literature analysis

Materials science literature analysis is notoriously difficult because it requires integrating composition, processing, characterization, and property relationships. Conventional retrieval-augmented generation (RAG) pipelines struggle with these heterogeneous tasks within a single retrieve-then-generate structure. Researchers led by Bixuan Li and Peng Kang have now developed AlphaAgent, a novel skill-contracted agent framework that explicitly separates retrieval-based question answering from paper-level report generation. The framework uses two dedicated skills: a retrieval skill that rewrites user requests into material-specific search intents and queries a curated index of over 300,000 papers from the Journal Citation Reports Metallurgy and Metallurgical Engineering category, automatically reformulating queries when initial evidence is insufficient; and a report-generation skill that parses full-text PDFs to produce structured per-paper analytical reports and cross-paper summaries.

In a blind evaluation on 40 materials science questions, half of which required deep analytical reasoning, AlphaAgent substantially outperformed a baseline system matched for underlying model, document index, and retrieval scale. The largest improvement was in mechanistic explanation and awareness of credibility boundaries—key capabilities for scientific rigor. These results demonstrate that explicit task separation, refined retrieval intent, and evidence-aware generation can significantly improve large-language-model-based literature analysis, making AlphaAgent a promising tool for materials researchers who need trustworthy, evidence-backed literature reviews and synthesis.