An autonomous AI Web Search and Information Retrieval Agent built for AMD ROCm hardware infrastructure. The system features a containerized FastAPI endpoint running a ReAct agent loop powered by vLLM (Qwen2.5-7B-Instruct) with Flash-Attention, an optimized Trafilatura and BM25/BGE-Small retrieval reranking layer to protect token budgets, and a deterministic verbatim-quote verification engine to completely eliminate hallucinations.