Fall Seminar Series 2025: Kamal Choudhary

Artificial intelligence (AI) investment has reached unprecedented levels, with U.S. private investment alone totaling $109.1 billion in 2024. The rapid infusion of AI into scientific research creates transformative opportunities for discovery but also exposes critical gaps in the data and computational infrastructure. Specifically, materials design increasingly relies on integrating heterogeneous computational tools, databases, experimental equipments, and machine learning models. However, orchestrating these resources into coherent, reproducible workflows remains a bottleneck. We introduce AGAPI (AtomGPT.org API), a domain-specific and yet versatile platform that enables agentic AI for materials research by coupling large language models with structured access to domain-specific datasets, tools, and models. AGAPI implements an autonomous architecture that plans, executes, and synthesizes multi-step workflows, spanning data retrieval, structure generation, property prediction, machine learning force fields, protein structure prediction, tight-binding inference, X-ray diffraction, microscopy image analysis, and inverse design. Benchmarks across representative tasks demonstrate that AGAPI can reduce researcher effort and failure modes compared to manual tool chaining. By explicitly combining AGAPI with agentic AI principles, this work establishes a scalable and trustworthy foundation for accelerating materials discovery and design.