Agentic Verification of Municipal Incident Reports through Grounded Dialogue and Multimodal Evidence
Year: 2026
Type: conference-paper
Abstract: Municipal call centers receive incident reports whose reliability ranges from urgent and accurate to vague, duplicated, mistaken, or deliberately false.We describe an agentic AI system for triaging non-emergency reports by combining adaptive questioning, consent-based location evidence, street-level imagery, public service-request history, and external news or official feeds.The system is framed around a New York City 311 pothole-reporting scenario, where a citizen calls about a road hazard that may require Department of Transportation action.Rather than treating a caller as truthful or deceptive from language alone, the system builds a grounded claim graph and estimates whether the report is actionable, uncertain, or likely abusive.The design emphasizes lowfriction questioning, evidence provenance, human review, and safety-first escalation when injury or imminent danger is possible.We argue that agentic verification can reduce prankcall load while preserving access for genuine reporters, provided it is deployed with privacy, bias, and accountability constraints.
Authors: Mohd Ariful Haque, Kishor Datta Gupta
Institution: Clark Atlanta University
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W7202293418