A Consensus-Driven Multi-LLM Pipeline for Missing-Person Investigations

Researchers Joshua Castillo and Ravi Mukkamala have introduced Guardian, a novel AI pipeline designed specifically for missing-person investigations, with a focus on the critical first 72-hour window. The system employs a multi-LLM architecture, coordinating several specialized large language models to perform intelligent information extraction and processing from case-related data. A key innovation is its consensus engine, which compares outputs from multiple models and resolves disagreements to produce more reliable results. The pipeline is further strengthened using QLoRA (Quantized Low-Rank Adaptation) fine-tuning on curated datasets, enhancing its accuracy for this sensitive domain.

Unlike approaches that treat AI as an unconstrained decision-maker, Guardian's design aligns with weak supervision principles, positioning LLMs as structured extractors and labelers. This creates an auditable, conservative system where human investigators remain in the loop, using the AI to rapidly synthesize information and plan early search operations. The paper, accepted to the CAC: Applied Computing & Automation Conferences 2026, details a 16-page framework with 6 figures, outlining how this pipeline could transform the initial, time-pressured phase of missing-child cases by providing investigators with a powerful, coordinated analysis tool.