Migrated from iterative LLM API call workflows to a tool-call-based agentic architecture, achieving 93% retrieval accuracy and pushing UAT ratings from 1.0 to 4.1 / 5.0 across enterprise evaluators. – Built multi-step reasoning and retrieval workflows for large technical document corpora (10,000+ documents) with streaming LLM responses and contextual memory management. – Implemented schema-aware retrieval tools, ACL-based search filtering, and citation-aware answer generation pipelines. – Participated in on-site deployment and infrastructure setup for local/offline enterprise deployment requirements.