Interpretable Markov-Based Spatiotemporal Risk Surfaces for Missing-Child Search Planning with Reinforcement Learning and LLM-Based Quality Assurance

Researchers Joshua Castillo and Ravi Mukkamala have introduced 'Guardian,' a novel AI system designed to assist in the critical first 72 hours of a missing-child investigation. The system addresses the challenge of fragmented, unstructured data by converting case documents into a schema-aligned spatiotemporal representation, enriching them with geocoding and transportation context. Its core innovation is a three-layer predictive architecture that generates actionable search plans. The first layer uses an interpretable Markov chain model that incorporates factors like road accessibility, seclusion preferences, and corridor bias, with separate parameters for day and night, to create probabilistic risk surfaces.

These risk distributions are then transformed into operationally useful search plans by a second layer powered by reinforcement learning (RL), which optimizes resource allocation. Finally, a third layer uses a Large Language Model (LLM) to perform post-hoc validation of the RL-generated plans before release, acting as a quality assurance check. The researchers validated the system using a synthetic but realistic case study, reporting quantitative outputs across 24, 48, and 72-hour horizons. The results demonstrate that Guardian produces interpretable priors for zone optimization, effectively bridging the gap between raw data analysis and actionable law enforcement strategy.