Spring AI RAG Multi-Topic System

Overview

Your Spring-AI-Topic-RAG project is a production-ready Retrieval-Augmented Generation (RAG) system built with Spring AI, Ollama, and Qdrant. Tthis project demonstrates an advanced implementation of AI-powered document processing and semantic search.

What You Built

Core Features:

• Multi-Topic RAG System: Separate, isolated RAG instances for different domains (Pentesting, IoT, Blockchain, Cloud, etc.)
• Semantic Search: Intelligent document retrieval using vector embeddings
• Document Processing: Support for PDF and Markdown files with automatic metadata extraction
• Local LLM Processing: Runs entirely locally using Ollama (no external API dependencies)
• Vector Storage: Efficient semantic search using Qdrant
• Easy Extensibility: Simple configuration to add new topics

Technology Stack:

• Framework: Spring Boot 3.5.8 with Spring AI 1.1.2
• Language: Java 21
• LLM: Ollama (local language model)
• Vector Database: Qdrant
• Document Processing: Apache Tika & PDFBox
• Communication: gRPC for Qdrant client
• Build Tool: Maven

---

Step-by-Step Explanation for Beginners

What is RAG (Retrieval-Augmented Generation)?

RAG is a technique that combines:

1. Retrieval: Finding relevant documents from a knowledge base
2. Augmentation: Using those documents to enhance the AI's response
3. Generation: Creating an answer based on both the user's question and the retrieved documents

Think of it as giving an AI assistant a library of books before asking questions!

How This Project Works - 5 Steps

Step 1: Document Ingestion

Code

Your PDF/Markdown Files
        ↓
    Tika & PDFBox (read files)
        ↓
    Extract Text & Metadata

• The system reads your documents and extracts text content
• Metadata (author, creation date, etc.) is automatically captured
• Documents are split into manageable chunks

Step 2: Vector Embedding

Code

Document Text
     ↓
Ollama (local LLM)
     ↓
Convert to Vectors (numerical representation)

• Each document chunk is converted into a vector (a list of numbers)
• These vectors capture the semantic meaning of the text
• Similar documents have similar vectors

Step 3: Vector Storage (Indexing)

Code

Document Vectors
     ↓
Qdrant Vector Database
     ↓
Organized & Searchable Index

• Vectors are stored in Qdrant for fast retrieval
• Organized by topic to keep domains separate
• Enables quick semantic search

Step 4: Query Processing

Code

User Question
     ↓
Convert to Vector (same way as documents)
     ↓
Search Qdrant for Similar Vectors
     ↓
Retrieve Top Matching Documents

• When a user asks a question, it's converted to a vector
• The system finds documents with similar vectors
• Only relevant documents are retrieved

Step 5: Response Generation

Code

User Question + Retrieved Documents
     ↓
Ollama (local LLM)
     ↓
Generate Intelligent Answer

• The LLM reads the retrieved documents
• It generates an answer grounded in those specific documents
• Response is more accurate and verifiable

Multi-Topic Architecture

Instead of one large database, you have separate RAG systems for different domains:

Code

Spring AI Application
    ├── Pentesting RAG ──→ Pentesting Documents → Pentesting Vector Store
    ├── IoT RAG ──────────→ IoT Documents → IoT Vector Store

Benefits:

• Better semantic relevance (avoids mixing unrelated domains)
• Faster searches (smaller databases)
• Easy to manage (add/remove topics independently)

Getting Started - Prerequisites

Before running this project, you need:

Requirement	Purpose
Java 17+	Run the Spring application
Maven	Build & manage dependencies
Docker & Docker Compose	Run containerized services
Ollama	Local language model (runs AI locally)
Qdrant	Vector database (stores & searches vectors)

Key Technologies Explained

Technology	Role
Spring Boot	Web framework for the application
Spring AI	Abstraction layer for AI/ML operations
Ollama	Runs large language models locally (privacy-first)
Qdrant	Specialized database for vector similarity search
Apache Tika	Extracts text from various document formats
PDFBox	Reads PDF files and metadata
gRPC	Fast communication protocol between services

Workflow Example

Scenario: You have PDFs about cloud security

1. Upload PDFs → System reads and chunks them
2. Index → Each chunk gets converted to vectors and stored in Qdrant
3. User Asks → "What are cloud security best practices?"
4. Search → Finds relevant chunks about cloud security
5. Generate → Ollama writes an answer based on those chunks
6. Return → User gets an accurate, sourced answer

---

Why This Project is Useful

✅ Private: Everything runs locally, no data sent to external APIs
✅ Accurate: Answers are grounded in your actual documents
✅ Customizable: Add any topic/domain you need
✅ Scalable: Separate topics mean independent scaling
✅ Modern: Uses cutting-edge Spring AI framework