I need an AI-powered chatbot that can sit on our website and handle real-time customer support queries by pulling accurate answers from our own company data. Visitors should be able to type natural-language questions about policies, services, and account issues and receive concise, helpful responses 24/7 without human intervention. Here is the scope I have in mind: • Core engine: a large-language-model setup (GPT-4, Claude, or a comparable open-source model) orchestrated through Python with LangChain or a similar framework. • Knowledge base: ingest PDFs, HTML pages, and structured documents from our internal drive; store embeddings in a vector database (Pinecone, Weaviate, or similar) so the bot retrieves only up-to-date company data. • Web deployment: a lightweight JavaScript or React widget that plugs straight into our existing site and streams answers back to the user. • Context management: prevent hallucinations, cite the source section used, and gracefully hand off to a human agent when confidence is low. • Admin console: simple dashboard where I can upload new documents and see usage analytics. Deliverables will be the fully commented source code in a Git repo, a brief setup guide I can hand to my dev-ops team, and a short video demo proving the workflow from question to answer on the live site. I’ll consider the project complete once the chatbot is live on our staging domain, consistently answers a predefined test suite with >90 % accuracy, and passes a one-week pilot with no critical failures. If this sounds like the kind of build you’ve tackled before, tell me how you’d approach the retrieval pipeline and which model you’d choose.