LLM Agent for Crash Analytics and Adaptive Feature Enrichment

Overview

• Builds an LLM-powered agent using LangGraph to analyze structured crash records.
• Enriches each case with context such as retail density, nightlife concentration, and other external signals.
• Generates analyst-facing crash case reports with likely contributing factors and supporting evidence.

Methods and contribution

The core idea is to turn raw crash records into richer analytical objects by combining the original structured data with external contextual signals that may explain traffic risk patterns more effectively than the base fields alone.

The agent is designed to adapt over time as conditions change, which supports continuous iteration without rebuilding the full modeling pipeline whenever roadway, policy, or seasonal patterns shift.

Materials