SCEC Award Number 24188 View PDF
Proposal Category Individual Research Project (Single Investigator / Institution)
Proposal Title SeismoGPT: A pilot Large Language Model (LLM) for seismology
Investigator(s)
Name Organization
Ahmed Elbanna University of Illinois at Urbana-Champaign
SCEC Milestones A1-2 SCEC Groups CCB, Seismology, RC
Report Due Date 03/15/2025 Date Report Submitted 03/30/2025
Project Abstract
The rapid evolution of large language models (LLMs) has enabled significant advancements in artificial intelligence (AI)-driven education and problem-solving. However, despite their access to vast datasets, existing models struggle with domain-specific challenges, particularly in engineering and scientific fields. This research explores the development, iterative refinement, and evaluation of multiple specialized LLMs tailored for complex analytical problem-solving. Initially, we developed StructurAI, an AI model for structural engineering, and SeismoSeed, a model focused on seismic studies. Through the deployment of these models, we encountered limitations inherent to proprietary LLMs such as GPT, particularly their closed-source nature, which restricted fine-tuning capabilities. Consequently, we reassessed various LLMs, benchmarking their performance, and ultimately selected DeepSeek’s open-source framework for fine-tuning and reconfiguration to enhance its data retrieval and question-answering capabilities for field specific inquires. Additionally, we evaluated the latest DeepSeek model alongside GPT-3.5, GPT-4.0, and Claude by testing them on subject-specific problems, analyzing their percentage errors, and compiling a comparative performance graph. This report outlines our methodology, initial findings, and future implications for domain-specific AI models.
Intellectual Merit This study demonstrates the effectiveness of a cyclical, iterative approach to developing domain-specific AI models for engineering and scientific problem-solving. By combining Retrieval-Augmented Generation, fine-tuning methodologies, and comparative benchmarking, we have developed StructurAI and SeismoSeed as specialized GPT tools, while also enhancing DeepSeek’s capabilities to provide better future models. Future work will focus on further refining these models, expanding their applications, and integrating them into real-world engineering and educational environments.
Broader Impacts The advancements in domain-specific LLMs have significant implications for AI-driven education, research, and industry applications. Personalized AI tutoring could revolutionize engineering education by providing adaptive learning experiences tailored to individual needs. Furthermore, AI-assisted analysis could streamline structural safety assessments and regulatory compliance, benefiting industries reliant on engineering expertise. Open-source contributions from this research will also provide valuable resources for the community, promoting further advancements in specialized AI models.
Project Participants Majd Alahmad : Undergraduate student in Civil and Environmental Engineering
Napat Tainpakdipat: PhD student in Civil and Environmental Engineering
Ahmed Ibrahim: PhD student in Civil and Environmental Engineering
Exemplary Figure Figure 1- Percentage Error in Structural Analysis Application across different LLM platforms
Linked Publications

Add missing publication or edit citation shown. Enter the SCEC project ID to link publication.