Although thousands of superconducting materials are reported in the literature, no single database captures the full breadth of experimental parameters beyond the critical temperature, Tc. Here, we present a new machine-assisted approach for data curation to construct a comprehensive database, AutoSuperCon [1]. We use an automated pipeline to extract diverse superconducting parameters from open-access preprints available on the arXiv server. Our three-stage workflow parses manuscripts into machine-readable text, segments and routes text to parameter-specific extractors via keyword matching, and uses a large language model for structured data extraction. The database contains multiple parameters, such as the critical temperature, upper critical field, and superconducting energy gap, for over 6,000 unique superconducting materials based on studies at both ambient and applied pressures. We demonstrate accurate data retrieval through close agreement with NIMS SuperCon [2] for the critical temperature and with curated reference values across other extracted parameters. AutoSuperCon complements existing theoretical databases of superconducting parameters [3,4] and provides a valuable resource for training machine learning models. Furthermore, AutoSuperCon enables large-scale statistical analysis of superconductor families. Specifically, we utilise these curated parameters to analyse deviations from the weak-coupling BCS gap ratio across different superconducting families. By systematically capturing experimental parameters, AutoSuperCon provides a robust foundation for testing microscopic pairing theories at scale and accelerating data-driven superconductor discovery.

[1] M. Manium et al., manuscript in preparation.
[2] NIMS SuperCon Database, MDR SuperCon Datasheet, https://doi.org/10.48505/nims.3739.
[3] B. Geisler et al., arXiv:2601.21044 (2026).
[4] T. F. T. Cerqueira et al., Adv. Mater. 36, 2307085 (2024).