Create a full-stack AI Research Assistant web application using React, Node.js, Express, and Tailwind CSS. Implement secure file uploading for PDF and plain text research documents, extracting raw text content for vector embedding generation. Connect to the OpenAI API to generate dense text embeddings and store vector embeddings in Supabase Vector or Pinecone for rapid semantic search retrieval. Build an interactive chat workspace where users can submit complex research questions and receive real-time streaming answers derived directly from uploaded documents, complete with inline source citations and page reference links. Include an automated web search fallback mechanism using Tavily search API whenever uploaded documents lack sufficient context. Add convenient export options allowing users to save conversation logs and synthesized research summaries as downloadable Markdown files or styled PDF reports. Design an intuitive multi-tab interface featuring document library management, prompt engineering selection presets, and real-time LLM token usage tracking statistics.