⚡ Gemini-Powered Persistent RAG System

A high-efficiency Retrieval-Augmented Generation (RAG) pipeline built with LangChain, FAISS, and Gemini 3 Flash. This system is engineered to handle local PDF documents while strictly respecting Google AI Studio's free-tier API quotas through persistent vector storage.

• Smart Persistence: Automatically saves/loads FAISS indices to disk to bypass daily embedding limits.
• Quota Optimized: "One-Hit" batching strategy to stay under 100 RPM and 1,000 RPD.
• Modern LLM: Powered by Gemini 3 Flash for fast, grounded reasoning.
• Error Resilient: Built-in auto-retry logic for 429 RESOURCE_EXHAUSTED errors.

Section: Environment Variables

To run this project, you will need to add the following environment variables to your .env file:

GOOGLE_API_KEY - Your Gemini API key from Google AI Studio

Section: Installation

Install the project dependencies with pip:

  python -m venv env 

  .\env\Scripts\activate

  pip install -r requirements.txt

Section: Run Locally

1. Ensure your PDF (e.g., `demo1.pdf`) is in the root directory.
2. Run the main script: python rag.py
3. On the first run, the script will generate a faiss_index_store folder. Subsequent runs will load this locally without using embedding quota.

---

Section: Tech Stack

Language: Python 3.9+  
Orchestration: LangChain  
LLM: Google Gemini 3 Flash  
Embeddings: Gemini-Embedding-001 (3072 dimensions)  
Vector Database: FAISS (Facebook AI Similarity Search)

Section: FAQ

When does my quota reset?

The 1,000 RPD (Requests Per Day) limit resets daily at 12:00 AM PT (12:30 PM IST).

How do I change the PDF?

Simply replace the PDF file in the directory and delete the faiss_index_store folder so the script knows to re-index the new data.

Does this work offline?

The vector retrieval works offline once the index is created, but the Gemini 3 Flash model requires an internet connection to generate the final answer.