Built an AI agent that reads incoming customer emails, finds the answer in the company's help docs using RAG (retrieval-augmented generation), and drafts a reply automatically.
The key feature: it scores its own confidence. When it's sure, it drafts a reply. When the docs don't cover the question, it escalates to a human instead of guessing so it never hallucinates answers to customers.
Stack: n8n, Gemini, vector embeddings for retrieval, confidence-gated routing. Cuts support response time while keeping a human in the loop for edge cases.