Platinum group metals (PGMs) underpin many catalytic technologies but face severe supply constraints, motivating the search for alternative materials and computational methods to accelerate discovery. While atomistic simulation tools such as Pymatgen and ASE have streamlined structure manipulation, they require detailed inputs, limiting accessibility for experimentalists and slowing early-stage exploration. Here, we present an AI-driven agentic framework that orchestrates worker–supervisor large language models (LLMs). The worker translates natural-language prompts of varying abstraction into valid crystallographic structures using a compact LLM fine-tuned with low-rank adaptation on a curated text–code–CIF dataset, emphasizing energy-efficient training. Benchmarking against the baseline CodeGen-350M-mono model shows that fine-tuning reduces hallucination rates from 100% to as low as 5% and improves structural match accuracy to up to 82% for fully specified inputs. Accuracy declines with decreasing prompt detail but remains nontrivial even when only stoichiometry and space group are provided, underscoring the LLM’s capacity for crystallographic inference. The supervisor Claude LLM evaluates the outputs and triggers iterative refinement through the worker’s built-in structure manipulation capabilities (e.g., supercell scaling, strain, vacancy, and substitution operations). We further demonstrate use cases for technologically relevant catalysts, including IrO 2 , pyrochlore Pb 2 Ir 2 O 7 , Ni 2 FeO 4 , and Ni 3 Mo , where the framework generates physically consistent structures that can be refined via geometry optimization. This work introduces a low-energy, language-driven pathway for integrating human and machine intelligence in materials design, paving the way for AI-assisted synthesis planning and high-throughput screening of complex oxides. • Compact AI agentic framework for crystal structure generation : a fine-tuned 350M-parameter worker model (CodeGen) and large supervisor model (Claude) successfully translate natural language into programmatic Pymatgen code, enabling usable crystal structures without billion-scale LLMs. • Energy-efficient and responsible AI : training and inference use orders of magnitude fewer resources than large foundation models, aligning with sustainable AI practices while retaining meaningful crystallographic reasoning. • Programmatic rather than file-based generation : the agent outputs Pymatgen code instead of full CIFs, making results compact, interpretable, and easily verified or manipulated by researchers. • Strong performance gains from fine-tuning : hallucination rates decreased from 100% in the baseline to as low as 5%, with structural match accuracy up to 82% for fully specified prompts. • Flexible handling of prompt abstraction : the model accommodates a spectrum of descriptions, from fully detailed crystallographic parameters to minimal inputs (stoichiometry + space group), demonstrating potential for both experts and non-experts. • Practical utility for catalysis : applied to benchmark noble-metal and earth-abundant catalysts (IrO 2 , Pb 2 Ir 2 O 7 , Ni 2 FeO 4 , Ni 3 Mo), the agent generated physically consistent structures suitable for iterative refinement. • Built-in structure manipulation capabilities : the agent can perform supercell scaling, strain application, vacancy creation, and substitution operations directly through natural language commands. • Outlook for integration : this work paves the way for multimodal AI agents that couple text-driven generation with simulation, property prediction, and closed-loop experimental design.
Baibakova et al. (Sun,) studied this question.