Customer Support LLM Chatbot

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.

Project ID: 40290250