LLM Augmented Financial Data Exploration via Programmatic Filter Synthesis

Traditional stock screening platforms present a significant barrier for non-technical users due to their reliance on com plex, quasi-programmatic syntax.This paper introduces a conversational financial analysis system that overcomes this limitation through a process termed "Programmatic Fil ter Synthesis" powered by a dual-LLM architecture for enhanced safety and reliability.A rigorous evaluation on a 100-prompt test suite demonstrated the system's capabilities, achieving an 87% semantically adjusted accuracy in translating natural language into structured S-expression filters.This work provides a robust blueprint for democratizing access to complex, data-driven financial tools by lever aging the power of conversational AI.