This research introduces a groundbreaking framework for automating materials synthesis by combining extensive data with Large Language Models that can judge synthesis quality at expert level.

• Created AlchemyBench: a comprehensive dataset of 17K expert-verified synthesis recipes from open-access literature
• Developed LLMs capable of evaluating synthesis procedures with expert-like precision
• Established a systematic approach to transform empirical trial-and-error methods into data-driven discovery
• Demonstrated potential to accelerate engineering innovations in energy storage, catalysis, and electronics

This advancement could dramatically reduce development time for new materials, enabling faster innovation cycles across multiple engineering disciplines from energy to biomedical applications.

Towards Fully-Automated Materials Discovery via Large-Scale Synthesis Dataset and Expert-Level LLM-as-a-Judge