Custom GPT with Gemini API (pass project) deploy on vercel

I’ve already sketched out a Custom GPT that lets a user plug in a personal Gemini API key, upload a PDF, and then question the model as though that uploaded text were part of its native knowledge. What I need now is a clean, production-ready build that keeps the Gemini integration front-and-center, handles PDF text extraction flawlessly, and returns concise, context-aware answers whenever a user asks something specific to their document.

Core flow
• User enters their Gemini key → immediate verification
• User uploads one or more text-based PDFs → content is parsed, chunked, and indexed
• User asks a personalized question → response is generated exclusively through Gemini, grounded on the stored PDF context

Key requirements
• Robust Gemini API usage with proper error handling and quota awareness
• Accurate PDF text extraction (PyPDF2, pdfplumber, or similar—open to your suggestion)
• Context retrieval that scales (vector store via FAISS, Pinecone, or a lightweight alternative)
• Secure front-end entry for the key and documents, preferably React or a minimal Flask UI
• Clear, commented code plus a short README so anyone can spin up the project locally

Acceptance criteria
1. A user can paste a valid Gemini key, upload a text-based PDF, and receive a relevant, citation-rich answer to a context-specific question.
2. If the key is invalid or the file is not textual, the system responds with an informative error.
3. All sensitive data (API keys, document content) stays local to the session—no third-party logging.

If you’ve built RAG (Retrieval-Augmented Generation) systems or fused Google Gemini into Python apps before, this should feel familiar. Send over a brief note on similar projects and the stack you’d like to use, and we can get started right away.