HAZDIAL

Structured agentic dialogue for hazard identification analysis.

Research question

Can multi-agent, multi-turn dialogue improve hazard identification over single-pass inference?

Research approach

The study compares adversarial debate and constructive discussion, with genetic algorithm-based interaction optimization. Evaluation considers classification quality and dialogue behavior.

Research outputs & focus

Dialogue-driven hazard analysis
Comparisons of debate and discussion
Hazard-identification evaluation

Related publications

The research behind the tool.

SIGDIALPublished

Enhancing Operational Safety via Agentic Dialogue Hazard Identification Analysis

Sanjay Das, Ran Elgedawy, Ethan Seefried, Ryan Burchfield, and Tirthankar Ghosal

Read paper ↗Explore HAZDIAL →

Citation

Sanjay Das, Ran Elgedawy, Ethan Seefried, Ryan Burchfield, and Tirthankar Ghosal (2026). Enhancing Operational Safety via Agentic Dialogue Hazard Identification Analysis. SIGDIAL, pp. 757–772.