Location
Havener Center, Miner Lounge / Wiese Atrium, 1:30pm-3:30pm
Start Date
4-1-2026 1:30 PM
End Date
4-1-2026 3:30 PM
Presentation Date
April 1, 2026; 1:30pm-3:30pm
Description
Technical documents, such as medical patents, are often inaccessible to non-expert audiences due to specialized terminology. This paper presents a multi-agent system for automated lay summarization that optimizes both quality and computational efficiency through strategic task decomposition. The proposed pipeline utilizes four specialized stages: autonomous medical entity retrieval from the Unified Medical Language System (UMLS), context distillation into lay conceptualizations, and a structured writer-critic loop for iterative refinement. By distributing cognitive load across specialized agents, this architecture effectively leverages smaller, cost-effective language models while maintaining the performance of SOTA LLMs. Evaluated against single-agent baselines using an LLM as a Judge and token usage tracking, the system demonstrates high-fidelity technical summarization at a fraction of the computational cost. These findings suggest that multi-agent task decomposition and domain specific RAG system offer a resource efficient path toward making dense technical literature accessible to simplified reading levels.
Biography
Manav is a Computer Science student at Missouri University of Science and Technology, expecting to graduate in December 2026. His academic interests lie at the intersection of Large Language Models and Agentic AI focusing on the implementation of AI systems for everyday use. He actively engages in research related to these areas, participating in the 2025-2026 OURE Program.
Beyond his primary research, Manav serves as president of the university's ACM (Association of Computing Machinery) student chapter where he fosters technical collaboration. Upon graduation, he intends to pursue a career in Industry to further advance the field of AI Engineering. He is dedicated to bridging the gap between computational innovation and its practical implementation for broader societal benefit.
Meeting Name
2026 - Miners Solving for Tomorrow Research Conference
Department(s)
Computer Science
Document Type
Poster
Document Version
Final Version
File Type
event
Language(s)
English
Rights
© 2026 The Authors, All rights reserved
Included in
Lay Summarization for Medical Patents
Havener Center, Miner Lounge / Wiese Atrium, 1:30pm-3:30pm
Technical documents, such as medical patents, are often inaccessible to non-expert audiences due to specialized terminology. This paper presents a multi-agent system for automated lay summarization that optimizes both quality and computational efficiency through strategic task decomposition. The proposed pipeline utilizes four specialized stages: autonomous medical entity retrieval from the Unified Medical Language System (UMLS), context distillation into lay conceptualizations, and a structured writer-critic loop for iterative refinement. By distributing cognitive load across specialized agents, this architecture effectively leverages smaller, cost-effective language models while maintaining the performance of SOTA LLMs. Evaluated against single-agent baselines using an LLM as a Judge and token usage tracking, the system demonstrates high-fidelity technical summarization at a fraction of the computational cost. These findings suggest that multi-agent task decomposition and domain specific RAG system offer a resource efficient path toward making dense technical literature accessible to simplified reading levels.
Comments
Advisor: Suman Maity, sm7rh@mst.edu