OrionTest: An Agentic AI Driven Automation*

Large language models are increasingly applied to software test generation, but existing approaches often exhibit non deterministic behavior and hallucinated user interface elements. This paper presents a deterministic multi agent architecture for requirement driven behavior driven development test synthesis. The system integrates structured retrieval, grammar constraints, and multi layer guardrails to ensure reproducible and context grounded outputs. Experimental results demonstrate improved consistency, reduced hallucinations, and stronger compliance with canonical test patterns compared to conventional stochastic generation methods.