This project demonstrates a fully local Retrieval-Augmented Generation (RAG) pipeline implemented in Java 21 with Gradle.
It works without any cloud dependencies — all processing happens locally, including embedding generation.

The system performs:

1. Exporting documents (example: Confluence REST API)
2. Cleaning HTML → plain text
3. Chunking documents
4. Generating embeddings locally (via a Python SentenceTransformers server)
5. Performing similarity search
6. Building an LLM prompt + querying an LLM endpoint

You can find my blog post about this example right here:

https://tuhrig.de/local-rag

• Java 21
• Gradle

pip install flask sentence-transformers

Optional if not already installed:

pip install torch

You also need a local embedding model:

📥 Download model (all-MiniLM-L6-v2):

https://www.kaggle.com/datasets/sircausticmail/all-minilm-l6-v2zip

Extract it into e.g.:

D:/embedding_models/all-MiniLM-L6-v2/

Then update the path in embedding_service.py:

MODEL_PATH = r"D:\embedding_models\all-MiniLM-L6-v2"

Inside your Python environment:

python embedding_service.py

You should see:

Loading model from: D:\embedding_models\all-MiniLM-L6-v2
Embedding server running on http://localhost:5005 ...

Your Java code will now POST requests to:

http://localhost:5005/embed

Below is the correct order to run all Java components.

new ConfluenceExporter().exportSpace("<YOUR_CONFLUENCE_TOKEN>");

new HtmlCleanerAndChunker(600, 100).process();

new ChunkEmbedder().embedAllChunks();

./gradlew run

Set these environment variables:

export AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/
export AZURE_OPENAI_DEPLOYMENT=gpt-4o
export AZURE_OPENAI_API_KEY=your-secret-key

This project gives you a full, local RAG pipeline:

✔ Extract internal documents
✔ Clean them
✔ Chunk them
✔ Embed locally
✔ Perform similarity search
✔ Send a prompt to your preferred LLM

Everything runs fully offline except the final LLM request.