Language-Native Materials Processing Design by Lightly Structured Text Database and Reasoning Large Language Model

Researchers have developed an AI framework that transforms materials synthesis procedures from unstructured narrative text into actionable, computable knowledge using large language models and structured databases. The system successfully optimized boron nitride nanosheet synthesis in three iterations, demonstrating AI's potential to accelerate complex materials discovery beyond traditional trial-and-error approaches.

This research addresses a fundamental bottleneck in materials science: synthesis procedures exist predominantly as narrative text scattered across papers and lab records, resistant to computational optimization. By converting procedural knowledge into a queryable, structured format while preserving causal logic, the researchers created a bridge between human expertise and machine reasoning that neither traditional databases nor unaugmented language models could achieve alone.

The framework's innovation lies in its hybrid approach—combining semantic matching, lexical search, and parameter-aware filtering with retrieval-augmented generation. Rather than treating synthesis as a black-box optimization problem, it reasons over procedural narratives, enabling failure diagnosis and protocol revision through iterative feedback loops. The boron nitride nanosheet case study is particularly significant because it involves multivariate constraints and path-dependent outcomes where transferring published protocols across laboratories typically fails. Convergence within three rounds represents dramatic acceleration of what normally consumes months of expert iteration.

For the materials science and chemistry sectors, this signals a shift from AI-as-assistant toward AI-as-active-planner in discovery workflows. Companies developing materials, pharmaceuticals, or specialty chemicals could substantially reduce time-to-market and experimental costs. The approach generalizes beyond materials—procedural optimization applies to process engineering, manufacturing, and complex synthetic workflows across industries. This work validates that domain-specific knowledge extraction from text, paired with structured reasoning, outperforms generic large language model assistance. The practical demonstration that dispersed literature evidence integrates with real experimental failures positions AI as a knowledge synthesis tool rather than merely an information retrieval layer.

• →AI framework converts unstructured synthesis narratives into computable knowledge using structured text databases and language model reasoning.
• →System achieved high-quality boron nitride nanosheet synthesis in three iterations versus typical months of expert-led trial-and-error.
• →Experience-augmented reasoning integrates multi-source literature with experimental failure modes to guide protocol refinement.
• →Hybrid approach combining semantic matching, lexical search, and parameter filtering outperforms generic large language model assistance.
• →Framework generalizes beyond materials science to any complex procedural domain with path-dependent, multivariate optimization problems.