I need to put our live-chat support on autopilot with an AI assistant that can do three things from day one: answer routine inquiries, handle complaints with the right tone, and serve up accurate product information. The assistant will sit inside the existing live-chat widget on our website, so the solution has to plug in cleanly through API or SDK, pull context from our knowledge base, and escalate seamlessly to a human when confidence drops or a customer requests it. Key deliverables • A conversational model—fine-tuned or rules-augmented—that understands our domain language and brand voice. • A knowledge-base ingestion pipeline that stays in sync with FAQ updates, policy docs, and the product catalogue. • Integration scripts for the live-chat platform plus a simple dashboard or config file for future tweaks. • Test suite and hand-over documentation showing how to retrain, monitor accuracy, and adjust escalation rules. Acceptance criteria • ≥90 % correct response rate across a test set of real chat logs. • Sub-2-second average reply time under normal load. • Clear human-handoff flow confirmed in staging. If you have hands-on experience with NLP, chatbot frameworks (Rasa, Dialogflow, or similar), and live-chat APIs, I’m ready to get started right away.