## 📌 Overview

An end-to-end, privacy-focused Retrieval-Augmented Generation (RAG) agent that runs 100% locally on system hardware. The system ingests external documents into a local vector database and enables an AI agent to perform contextual document search—with zero cloud costs and zero data leaving the local machine.

---

## 🛠️ Tech Stack & Architecture

- **Workflow Orchestration:** n8n (Docker containerized)

- **Vector Database:** Qdrant Vector Store

- **Local LLM Engine:** Ollama (`llama3.2` for logic, `nomic-embed-text` for vector embeddings)

- **Container Infrastructure:** Docker Compose (with GPU acceleration)

- **Exposing Endpoints:** ngrok tunneling

---

## ⚙️ Key Technical Features & Workflow Breakdown

1. **Document Ingestion & Indexing Pipeline:**

  - Ingests text documents via default data loaders.

  - Splits content into uniform chunks using a `Token Splitter`.

  - Generates vector embeddings via local Ollama embedding models (`nomic-embed-text`).

  - Stores and indexes embeddings inside `Qdrant` vector memory.

2. **AI Agent Query & Retrieval Pipeline:**

  - Connects an interactive `AI Agent` powered by `llama3.2`.

  - Integrates `Qdrant Vector Store` directly as a tool available to the agent.

  - Dynamically searches indexed knowledge to answer complex technical queries accurately based on context.

---

## 🟢 Results & Key Takeaways

- **100% Local Processing:** Full data privacy achieved without sending data to third-party LLM APIs.

- **Cost Efficiency:** Zero API overhead or recurring subscription costs.

- **Modular Design:** Scalable pipeline adaptable to multiple document types or larger local LLM models.