Agentic Architecture: Architected a real-time AI Support Copilot using Next.js and Node.js, leveraging a LangGraph-orchestrated Llama 3.3 agent to automate enterprise knowledge retrieval across dense internal documents. Hybrid Retrieval: Engineered a highly accurate RAG pipeline in PostgreSQL (pgvector), combining semantic vector search with keyword matching (websearch to tsquery) and Reciprocal Rank Fusion (RRF) to maximize precision. Asynchronous Processing: Designed a fault-tolerant background processing engine using Redis and BullMQ to asynchronously ingest massive PDFs, utilizing calculated text overlaps to completely prevent semantic data loss without blocking the main server thread. Latency Optimization: Eliminated LLM hallucinations via strict agentic guardrails for deterministic answering and established Server-Sent Events (SSE) for real-time token streaming to minimize perceived UI latency