AI in Public Safety
Interpretable Markov-Based Spatiotemporal Risk Surfaces for Missing-Child Search Planning with Reinforcement Learning and LLM-Based Quality Assurance
This paper presents Guardian, an end-to-end decision-support system for missing-child investigations. It leverages a three-layered predictive system: a Markov chain for mobility forecasting, reinforcement learning for search plan optimization, and an LLM for quality assurance. The system converts unstructured case data into actionable search products, providing interpretable risk surfaces and search zones for law enforcement agencies. It emphasizes the critical first 72 hours of an investigation and aims to improve search planning accuracy and efficiency under uncertainty.
Unlocking Enhanced Search Efficiency with AI
Guardian redefines missing-child search operations by integrating advanced AI, reducing time-to-discovery, and optimizing resource allocation across critical phases.
0
Critical Hours Focused
0
Reduction in Search Area (Est.)
0
Improvement in First-Hit Accuracy (Pilot)
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
System Overview
Predictive Layers
Key Metrics
The Guardian system is a modular ecosystem comprising the Parser Pack (data preparation) and the Core System (analysis & evaluation). It processes raw, unstructured case documents into a schema-aligned spatiotemporal representation, enriching cases with geocoding and transportation context. It then produces probabilistic search products spanning 0-72 hours, designed to be auditable and consumable by investigators.
Guardian's core predictive component is a three-layer architecture. The first layer uses a Markov mobility forecasting component to produce sequential horizon forecasts. The second layer employs reinforcement learning to convert belief maps into actionable search zones. The third layer utilizes an LLM for post hoc validation and quality assurance of the search plans before release.
Performance is measured by Sector mass (%), indicating aggregated probability mass in operational sectors; Containment radii for 50/75/90% probability around the Initial Planning Point (IPP); and Hotspot concentration, quantifying cumulative probability mass captured by top-K hotspots. Illustrative metrics include Geo-hit@K and area-searched-until-hit.
72 Hours
Critical Window for Missing-Child Investigations
Guardian's 3-Layered Predictive Process
Markov Mobility Forecasting
→
Reinforcement Learning for Zone Optimization
→
LLM-Based Quality Assurance
| Feature | Traditional Approach | Guardian (AI-Driven) |
|---|---|---|
| Data Input |
|
|
| Prediction |
|
|
| Actionability |
|
|
| Interpretability |
|
|
| Adaptability |
|
|
GRD-2025-001541: A Synthetic Case Study
The system was evaluated using GRD-2025-001541, a synthetic yet realistic case of a 15-year-old female last seen in York, Virginia. Forecasts were generated at 24h, 48h, and 72h horizons. Results showed strong concentration in the Tidewater region (50%+ probability), with Northern Virginia emerging as a secondary region due to corridor connectivity. Spatial uncertainty expanded in a structured manner over time, preserving corridor-aligned structures, demonstrating the model's ability to provide interpretable priors for zone optimization and human review.
Calculate Your Potential Savings
See how much time and cost your organization could reclaim by integrating Guardian's advanced AI capabilities into your operations.
Annual Cost Savings
$0
Annual Hours Reclaimed
0
Your Path to AI-Powered Search
A structured timeline for integrating Guardian into your public safety operations, ensuring a smooth transition and measurable impact.
Phase 1: Discovery & Data Integration
Initial assessment, data source identification, and integration with Guardian Parser Pack for schema alignment and enrichment.
Phase 2: Model Training & Customization
Calibration of Markov models with historical data, fine-tuning of RL parameters, and LLM training for case-specific QA.
Phase 3: Pilot Deployment & Validation
Deployment in a controlled environment, validation against real-world scenarios, and iterative feedback for refinement.
Phase 4: Full-Scale Rollout & Continuous Optimization
Operational deployment, ongoing monitoring, performance analysis, and adaptive adjustments for sustained efficiency.
Ready to Transform Your Search Operations?
Schedule a personalized consultation with our AI experts to explore how Guardian can enhance your team's effectiveness and reduce critical response times.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat