Oculi Automates Credit Risk Analysis with Conversational AI

Financial institutions typically rely on analysts to manually perform credit risk analysis, a process involving writing SQL queries, running statistical computations, and building visualization dashboards. This traditional workflow is time-consuming and often limits exploration to familiar data segments. A new conversational platform named Oculi aims to revolutionize this by automating the entire process. Oculi allows analysts to pose natural language questions, which the platform then converts into comprehensive credit risk analyses. It handles data queries, statistical testing, and generates interactive visualizations. The platform features a three-layer architecture that separates reasoning (handled by an LLM-powered agent), execution (via Model Context Protocol tool servers), and presentation (through an agentic UI). This design enables analysts to efficiently discover high-risk portfolio segments. A novel segment discovery pipeline within Oculi combines deterministic statistical methods with LLM-guided feature selection. This approach leverages the LLM's semantic domain knowledge alongside data-driven metrics to identify meaningful and actionable portfolio segments. When evaluated on a mortgage portfolio with over 200 features, Oculi demonstrated its effectiveness in uncovering material risk segments that were previously difficult to find manually, drastically reducing the time required for insights while maintaining auditability and statistical rigor.