Tag
This paper presents a synthetic multimodal framework for insurance fraud detection at the first notice of loss (FNOL). It generates dialogue transcripts and two-speaker audio, combining ASR, NER, LLM-RAG, and speaker embeddings into a rule-based risk scoring system.