LLM-Based AI Agents for Automated Data Extraction of Material Properties and Structural Features

The rapid discovery of materials is constrained by the lack of large, machine-readable datasets that couple performance metrics with structural context. Existing databases are either limited in scale, manually curated, or biased toward idealized first principles results, leaving experimental literature underexploited. Here we present an agentic, large language model (LLM) driven workflow that autonomously extracts thermoelectric and structural properties from ∼10,000 full-text scientific articles. The system integrates dynamic token allocation, zero-shot multi-agent extraction, and conditional table parsing to balance accuracy against computational cost. Benchmarking on a manually curated set of 50 papers shows that GPT-4.1 achieves the highest extraction accuracy (F1 ≈ 0.91 for thermoelectric properties, F1 ≈ 0.838 for structural fields), while GPT-4.1 Mini offers nearly comparable performance (F1 ≈ 0.889 for thermoelectric, F1 ≈ 0.833 for structural) at a fraction of the cost, enabling large-scale deployment. Applying this workflow, we create a data set of 27,822 property temperature records with normalized units, spanning the figure of merit (ZT), Seebeck coefficient, electrical conductivity, power factor, and thermal conductivity, together with structural attributes such as crystal class, space group, and doping strategy. To facilitate community access, we release an interactive web explorer supporting semantic filters, numeric queries, and CSV export. Together, this study delivers the largest LLM-curated thermoelectric dataset to date, provides a reproducible and cost-profiled extraction pipeline, and establishes a foundation for scalable and data-driven thermoelectric material discovery. The workflow is broadly generalizable and can be adapted to curate structure-property datasets across diverse classes of functional materials beyond thermoelectrics.