Answers are grounded in this paper and cite their sources.
Automated creation of actionable chemical datasets through large language model (LLM) interpretation of chemistry publications has become an engineering challenge rather than a frontier AI problem. We present Chemstractor , a bespoke extraction pipeline designed for use with open‑weight LLMs that can be executed locally on hardware typically available to university research environments. As a test case, the extraction techniques are applied to polymer diffusion coefficients, commonly used for the prediction of polymer molecular weights and for evaluating emerging universal calibration curves. The pipeline combines modern PDF parsing, table focused chunking, structured prompting, and an agent orchestrated calculator tool to handle mathematical transformations and reduce hallucinations. Chemstractor achieves strong performance, including a low hallucination rate, while maintaining extremely low token usage and requiring no paid API access. The methodology provides an accessible and efficient route to generating validated, analysis ready datasets and offers a template for extracting similarly complex scientific parameters in other domains.
No takes yet. Share an insight, caveat, or question.
Varley et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: