🧠 Vectorless RAG

A RAG (Retrieval-Augmented Generation) system built without any vector database or embeddings — using Google Gemini + PageIndex instead.

---

📌 What is Vectorless RAG?

Traditional RAG pipelines follow this pattern:

Chunk documents → Embed into vectors → Store in Vector DB → Query by similarity search

This project challenges that assumption. Instead of embedding-based retrieval, it uses PageIndex to index document pages directly. At query time, Gemini reasons over the indexed pages to find and synthesize relevant information — no embeddings, no ANN (approximate nearest neighbor) lookup.

---

🗂️ Project Structure

vectorless_rag/
├── PageIndex_Vectorless_RAG.ipynb   # Main notebook
├── attention_all_you_need.pdf       # Sample document (Attention Is All You Need)
└── .gitignore

---

⚙️ How It Works

1. Load a PDF document (e.g., Attention Is All You Need paper)
2. Index pages using PageIndex — no embeddings generated
3. Query using natural language
4. Gemini reasons over the indexed pages and returns a synthesized answer

---

✅ Advantages

• No vector DB setup or maintenance overhead
• No embedding model needed — reduces infra cost and latency
• Simpler pipeline: fewer moving parts = easier to debug
• Works well for structured/paginated documents (PDFs, reports)
• Faster to prototype and deploy
• Google Gemini is free to use 😂

---

⚠️ Limitations

• Scales differently — retrieval is not O(log n) like ANN search
• Less effective for large unstructured corpora where semantic similarity shines
• Dependent on Gemini's reasoning quality for relevance judgment
• PageIndex is optimized for document-style inputs; less flexible for arbitrary text chunks
• Not a drop-in replacement for vector RAG in all use cases

---

🚀 Getting Started

Prerequisites

• Python 3.12
• Google Gemini API key (free at Google AI Studio)
• Jupyter Notebook

Installation

git clone https://github.com/spylovec2/vectorless_rag.git
cd vectorless_rag
pip install -U pageindex google-generativeai python-dotenv

Set your Gemini API key

GEMINI_API_KEY    = os.getenv("GEMINI_API_KEY")

---

🛠️ Tech Stack

Tool	Purpose
Google Gemini	LLM for reasoning & answer generation
PageIndex	Document page indexing without vectors
Jupyter Notebook	Interactive development environment
Python	Core language

---

📄 Sample Document

The repo includes the "Attention Is All You Need" paper (attention_all_you_need.pdf) as a sample document to test the RAG pipeline against.

---

🙏 Credits

• Inspired by Krish Naik's video on RAG pipelines
• Sample document: Attention Is All You Need — Vaswani et al., 2017

---

📬 Connect

If you find this interesting or have feedback, feel free to open an issue or reach out on LinkedIn.

---

⭐ Star this repo if you found it useful!