Designed and implemented a multi-agent AI system for business stress testing, orchestrating specialized LLM agents (Pessimist, Resilience, Synthesizer) with real-time web intelligence to simulate, challenge, and synthesize high-fidelity risk scenarios. Built the end-to-end pipeline integrating structured prompting, Tavily-based search grounding, and JSON-normalized outputs to generate actionable, company-specific risk assessments and recovery strategies.