RAGlaw Project Documentation

Overview

This project is RAG model that embedds Polish's Penal Code and answers questions with gpt-4o-mini LLM model. It's deployed on the Azure Kubernetes Service (AKS)

Architecture

Main components

rag-api

The 'rag-api' is API written in FastAPI that is serving as a backend service for the RAG communication. It consists of one POST Endpoint that retrieves the relevant context in Milvus DB, feeds it into LLM and returns the LLM answer based on that context

frontend

The 'frontend' is streamlit application that communicates with the rag-api and provide user-friendly interface to chat with RAG model

ingesting

The 'ingesting' is Argo Workflows's workflow that chunks Polish's Penal Code, embedds it and stores the embeddings in Milvus Vector Database

kubernetes

Application is designed to run on kubernetes' pods

terraform

Azure Kubernetes Cluster (AKS) is deployed using terraform

Getting Started

Prerequisites

• Terraform installed
• Azure CLI installed
• kubectl installed
• Helm installed
• valid Azure subscription

Deployment Steps

1. Terraform Deployment

   • Navigate to the terraform directory.
   • Create terraform.tfvars file and assign values

     resource_group_name     = "law-rag-model-rg"
     resource_group_location = "West US"
     aks_name                = "law-rag-model-aks"
     dns_prefix              = "law-rag-model-dns"
     node_count              = 1
     vm_size                 = "Standard_B2s"
     os_disk_size_gb         = 32
     subscription_id         = "<YOUR_SUBSCRIPTION_ID>"

   • Initialize Terraform:

     terraform init

   • Apply the Terraform configuration to deploy the AKS cluster:

     terraform apply

2. Application Deployment

   • Navigate to the deploy directory.
   • Use Helm to deploy the application:

     helm install <release-name> ./helm-chart

   • Alternatively, apply the Kubernetes YAML files:

     kubectl apply -f sample.yaml -n namespace