RAG-based application for Q&A with data from your Confluence Wiki using LangChain, ChromaDB, AWS Bedrock or Ollama.
⚠️ The project is in a very early development stage. I would even call it as experimental, so don't rely on further development, support etc. However, we welcome contributions.
🇩🇪 The code is written in english but the prompt templates are currently only available in german, as the app is initially designed to be used in a german company. Feel free to submit PRs for other languages.
Before you start, those requirements should be met:
- Python installed
- Docker and docker-compose installed
- access to either Amazon Bedrock or Ollama embeddings and language models
- You should have a Personal Access Token (PAT) for Confluence
- if you want to use Bedrock, configure the credentials as described here
Set up a virtual invironment and install requirements.
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtIndexing is done by running the indexing.ipynb notebook. Before you run it, ensure that you environment is set up (see notebook) and that you have ChromaDB up and running. To run an instance locally, use the docker-compose.yaml provided in the repo. Run it with:
docker-compose up -dWhen run locally, Chroma will save it's data in the data/chroma directory.
Execute main.py from command line. Before you do that, make sure that you modify the CHUNK_SIZE constant at the top of the script according to the size you have chosen in the previous step. When the script is executed, you will be prompted to select a language model. The options are partially based on the Ollama models you have downloaded, if there are any. The other option is Amazon Bedrock.
Here are some useful resources related to RAG, LLMs etc.:
- LangChain Hub (for prompt templates)
- Playlist "RAG from Scratch by LangChain"
- basic RAG tutorial by pixegami (youtube)
- Amazon Bedrock Prompting Examples & Tools
- RAG using LangChain with Amazon Bedrock Titan text, and embedding, using OpenSearch vector engine
TODO