RAG Demo

Parallel implementations of Retrieval-Augmented Generation (RAG) in Java (Spring AI), Java (LangChain4j), and Python (LangChain) for teaching purposes.

All demos follow the same flow: Load PDF → Chunk → Embed → Store → Retrieve → Generate

Choose Your Implementation

	Java (Spring AI)	Java (LangChain4j)	Python
Framework	Spring AI 1.0.3	LangChain4j 1.10.0	LangChain
LLM	OpenAI (gpt-5-mini)	OpenAI (gpt-5-mini)	OpenAI (gpt-5-mini)
Embeddings	text-embedding-3-small	text-embedding-3-small	text-embedding-3-small
Vector Store	SimpleVectorStore (in-memory)	InMemoryEmbeddingStore	InMemoryVectorStore
PDF Parser	PagePdfDocumentReader	Apache Tika	PyPDFLoader
IDE	IntelliJ IDEA	IntelliJ IDEA	PyCharm / VS Code

Prerequisites

• OpenAI API Key — Set as environment variable OPENAI_API_KEY
• Java 21+ (for Java versions)
• Python 3.10+ (for Python version)

Quick Start

Java (Spring AI)

cd springai
export OPENAI_API_KEY=sk-...
./gradlew bootRun

Java (LangChain4j)

cd langchain4j
export OPENAI_API_KEY=sk-...
./gradlew run

Python (LangChain)

cd python
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
export OPENAI_API_KEY=sk-...
python -m ragdemo.main

Production Track (Supabase + pgvector)

This main branch focuses on a minimal, in-memory setup for teaching core RAG ideas quickly.

For a persistence-backed, production-style version across all three implementations (Spring AI, LangChain4j, and Python), use the supabase-pgvector branch:

git switch supabase-pgvector

That branch README includes full setup and run instructions for Supabase/pgvector.

Mode	Vector Storage	Setup Effort	Best For
Main branch	In-memory stores	Low	Learning the RAG pipeline mechanics
supabase-pgvector branch	PostgreSQL + pgvector (Supabase)	Medium	Persistence, realism, and deployment discussions

Project Structure

ragdemo/
├── springai/                        # Spring AI implementation
│   ├── build.gradle.kts
│   └── src/main/java/edu/trincoll/ragdemo/
│       ├── RagDemoApplication.java      # Spring Boot entry point
│       ├── RagDemoRunner.java           # Interactive CLI
│       ├── config/RagConfig.java        # VectorStore + ChatClient beans
│       └── service/
│           ├── DocumentLoaderService.java   # PDF loading + chunking
│           └── RagService.java              # Q&A via ChatClient
│
├── langchain4j/                     # LangChain4j implementation (no Spring)
│   ├── build.gradle.kts
│   └── src/main/java/edu/trincoll/ragdemo/
│       ├── RagDemo.java                 # Plain Java CLI entry point
│       ├── RagService.java              # RAG pipeline + InMemoryEmbeddingStore
│       └── DocumentLoader.java          # PDF loading via Apache Tika
│
└── python/                          # Python/LangChain implementation
    └── src/ragdemo/
        ├── main.py              # CLI entry point
        ├── document_loader.py   # PDF loading + chunking
        ├── vector_store.py      # Embedding + storage
        └── rag_chain.py         # LCEL chain (retriever → prompt → LLM)

How RAG Works

┌─────────────────────────────────────────────────────────────────┐
│                        INDEXING PHASE                           │
├─────────────────────────────────────────────────────────────────┤
│  PDF  ──▶  Load  ──▶  Chunk  ──▶  Embed  ──▶  Vector Store      │
└─────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────┐
│                        QUERY PHASE                              │
├─────────────────────────────────────────────────────────────────┤
│  Question ──▶ Embed ──▶ Search ──▶ Retrieve ──▶ Augment ──▶ LLM │
└─────────────────────────────────────────────────────────────────┘

1. Load — Extract text from PDF documents
2. Chunk — Split into smaller pieces (~800 tokens) for better retrieval
3. Embed — Convert text chunks to vector embeddings
4. Store — Save embeddings in vector store for similarity search
5. Retrieve — Find chunks most similar to the user's question
6. Generate — Send question + retrieved context to LLM for answer

Sample Questions

The demo includes the "Attention Is All You Need" paper. Try asking:

• What is the Transformer architecture?
• How does self-attention work?
• What are the key contributions of this paper?
• What is multi-head attention?

Adding More Documents

Drop additional PDF files into the documents/ folder:

• Java (Spring AI): springai/src/main/resources/documents/
• Java (LangChain4j): langchain4j/src/main/resources/documents/
• Python: python/documents/

The application will automatically load all PDFs on startup.

Running Tests

Java (Spring AI)

cd springai
./gradlew test

Java (LangChain4j)

cd langchain4j
./gradlew test

Python

cd python
pip install -e ".[dev]"
pytest

License

MIT License — See LICENSE for details.

Author

Kenneth Kousen