NexusMesh Guard is an auditable, multi-agent AI claims fraud detection and compliance platform designed for the auto insurance sector. With insurance fraud costing consumers and carriers over $308 billion annually, the industry requires advanced, explainable detection. NexusMesh replaces opaque, "black-box" rules engines with a fully transparent, 6-agent hybrid architecture coordinated via the Band SDK and powered exclusively by the AI/ML API. Multi-Agent Swarm Architecture The platform operates on a parallel-to-serial stateful flow utilizing six specialized agents: Intake Agent (Gemini-2.5-Flash): Rapidly ingests CSVs/PDFs, performs OCR, and streams structured FNOL data. Document Authenticity Agent (Qwen3.5-Omni-Plus): Executes deepfake detection using EXIF, C2PA, Error-Level-Analysis, and Vision-LM checks on damage photos. Fraud Detection Agent (MiniMax-M3): Uses graph-clustering to detect multi-claim fraud rings (e.g., shared tow companies) based on historic SIU outcomes. Regulatory Browser Agent (Grok-4-3): Leverages real-time, headless web search to fetch state compliance mandates and DOI bulletins. Policy Risk Analyzer (GPT-5.1): Parses dense ISO policy PDFs to ensure compliance with state minimums (e.g., Florida PIP requirements). Decision & Governance Agent (GPT-5-2): Aggregates findings behind a barrier, routing claims (Green/Yellow/Red). Crucially, it triggers Human-in-the-Loop (HITL) escalations for RED flags and generates NAIC-compliant reports. FACTS Compliance & Production Path To align with the NAIC AI Model Bulletin, NexusMesh implements a transparent FACTS layer (Fairness, Accountability, Compliance, Transparency, and Safety), including cryptographic logging of decisions and automated disparate-impact tracking. Designed as an enterprise overlay, it seamlessly integrates into Guidewire ClaimCenter or Duck Creek via REST APIs.
Category tags: