CASSIE: A Specialized AI Conversational Agent for Chronic Pelvic Pain Management ID 3638

INTRODUCTION: Chronic pelvic pain (CPP) affects 15–26% of the female population worldwide and, although common, is not well understood. Causes are complex and multifactorial, requiring specialized knowledge for management. General AI chatbots provide broad medical information but lack expertise. We investigated whether a specialized conversational agent could demonstrate superior accuracy and clinical relevance compared to general-purpose large language models (LLMs) in CPP management. METHODS: We developed CASSIE (Conversational Agent for Supportive Solutions and Information about Endometriosis) with deep-domain expertise in CPP management. A PubMed literature review using targeted search terms identified 278 publications from the past 25 years. After applying inclusion criteria (English, gynecological CPP focus, evidence-based management), 46 publications were selected and summarized into 300-word entries. CASSIE was built using base LLM architecture with a curated knowledge base and guardrail protocols. Performance evaluation used AMBOSS medical questions, comparing CASSIE with general-purpose LLMs. RESULTS: Preliminary evaluation demonstrated 82% and 94% accuracy using Claude and Gemini as base LLMs, respectively. Comparative testing revealed: ChatGPT (92%), Perplexity (94%), Claude (90%), and Gemini (86%). CASSIE exhibited strong CPP knowledge and adaptive learning capabilities, successfully synthesizing multidisciplinary literature into evidence-based responses. When informed of incorrect answers, CASSIE demonstrated error correction ability. CONCLUSIONS/IMPLICATIONS: CASSIE demonstrates the potential of specialized AI training for medical applications. Clinical benefits include health care provider decision-making support, standardization of management, and improved patient education. This establishes a foundation for specialized medical AI agents addressing critical knowledge gaps in underserved domains, with future directions including expanded literature integration and user experience studies.