Website Product AI Chatbot

I need a context-aware assistant that can live on our website and confidently answer product-specific questions. Your task is to build a Retrieval-Augmented Generation (RAG) agent—whether you prefer Voiceflow, Botpress, LangChain + LlamaIndex, or Flowise—then train it on our existing PDFs (product manuals, release notes, and past support threads exported to PDF).

The finished bot should:

• Parse those PDFs into a vector store (Pinecone, pgvector, or Chroma—your call)
• Deliver accurate responses in natural language, recognizing when to escalate to a human agent
• Log every conversation for analytics and continual improvement
• Embed smoothly as a React / Next.js widget styled with Tailwind

Acceptance criteria
1. Answers 90 %+ of test prompts pulled from real tickets without hallucinations.
2. Fallback flow triggers when confidence drops below a threshold we define together.
3. Deployment URL and simple instructions for my dev team to replicate in staging.

Please outline your preferred toolchain, prior RAG work, and any extra data you may need so we can start right away.