The rapid growth of scientific literature in materials sciences and other domains presents a growing challenge where critical process knowledge remains embedded in unstructured text, limiting its reuse for data-driven discovery. Agentic workflows, which integrate large language models (LLMs) with different tools, such as application programming interfaces (APIs), and rule-based functions, have emerged as a promising paradigm to overcome these limitations. In this work, we introduce SciKGExtract, an agentic artificial intelligence (AI) workflow for structured knowledge extraction from scientific publications. The workflow combines LLM-based contextual understanding with deterministic tools for schema validation, data cleaning, and normalization, leveraging authoritative external repositories such as PubChem to ensure accuracy and semantic consistency. Applied to atomic layer deposition (ALD) literature from the AtomicLimits database, SciKGExtract produces a structured dataset capturing experimental parameters and material properties across test cases of zinc oxide (ZnO) and indium-gallium-zinc oxide (IGZO). The extracted data reveal dominant ALD methods and chemistries, variability in reported growth-per-cycle and temperature ranges, and the need for richer metadata to enable reliable cross-study comparisons. These results demonstrate how structured extraction can transform unstructured literature into AI-ready knowledge, accelerating validation, benchmarking, and discovery in emerging materials systems.
Sadruddin et al. (2026) studied this question.