CarePal: Multilingual Contradiction-Aware Conversational System for Structured Symptom Intake

CarePal is a conversational intake system designed to collect symptom histories in settings where linguistic diversity, inconsistent narratives, and limited clinical time reduce the quality of pre-diagnosis information. The system integrates adaptive question generation, multilingual alignment, and the detection of contradictions through three core algorithmic components: (1) an Adaptive Question Generation Engine that dynamically constructs symptom inquiries based on patient responses, (2) a Multilingual Semantic Alignment Module that preserves meaning across Hindi, Bengali, and English, and (3) a Contradiction Detection and Resolution Framework that identifies and resolves inconsistencies in patient narratives. This paper outlines the architecture, training process, hyperparameter tuning, detailed algorithm specifications, and evaluation using quantitative metrics on 1,260 synthetic dialogues, providing simulation-based evidence of system performance, alongside usability data from a survey-based study with 26 participants. We hypothesized that CarePal would increase symptom-history completeness and internal consistency compared with a template-based intake system.