Welcome to our custom RAG (Retrieval Augmented Generation) project! We've crafted this using a micro-services architecture to bring you a powerful, yet flexible, solution.
Our project is composed of three key services 🛠️
-
Embedding API This service turns your text into numerical vectors, crucial for our machine learning magic! You have the flexibility to choose between:
- Sentence-transformers embedding models from Hugging Face.
- OpenAI Embeddings for advanced performance.
-
Knowledge Retriever Think of this as the brain of our project. It can:
- Extracts text from various documents (PDF, Word, PPTX) using the versatile unstructured Python library (also capable of OCR).
- Manages a 'vectorstore' for efficient similarity searches, powered by ChromaDB.
- Connects to Azure OpenAI GPT and Vertex AI Palm Text Generation Models, bringing cutting-edge AI into play.
-
Front App A user-friendly chat interface built with Gradio, making it easy for you to interact with our service.
Here's how to get everything up and running 🚀
-
Installation Process Simply build and run the images using our Docker compose file. Execute this command from the project's root directory: docker compose up --build
-
Software Dependencies Our Embedding & Knowledge Retriever services use a PyTorch image from DockerHub. For local testing, ensure that the torch library is installed in your Python environment.
-
API References To extend and enrich our services, we're leveraging: - GCP Vertex AI services - Azure OpenAI Services
The application is Kubernetes-ready! 🌟
-
Kubernetes Deployment You'll find everything you need in the Kubernetes Folder within the project. This includes all the necessary manifests to smoothly deploy the application on a Kubernetes cluster.
-
Continuous Delivery Made Easy
In our Pipelines Folder, we have set up a streamlined continuous delivery process using Azure DevOps Pipelines. There's a dedicated pipeline for each of our three services, and each pipeline includes two key stages:
- Build and Push: This stage takes care of building each service's image and pushing it to a Container Registry. It ensures that your latest code is always ready to be deployed.
- Update and Sync: In this stage, the Kubernetes manifests are updated with the commitID. This is crucial for ArgoCD, as it allows for synchronization with the latest image pushed to the container registry.
