LAB515: Build an Agentic App with PostgreSQL, GraphRAG, and Agent Framework

Overview

This lab guides you through developing an agent-driven, Retrieval-Augmented Generation (RAG) application that explores U.S. Case Law data. You'll learn how to combine PostgreSQL's powerful database capabilities with AI techniques to create a legal research assistant capable of providing accurate and contextually relevant answers.

Architecture

What You'll Build

• A Microsoft Agent Framework Agent that can reason over legal cases stored in PostgreSQL
• A system that utilizes vector embeddings for semantic search
• AI function tools for database search and external data retrieval
• An application enhanced with the GraphRAG pattern for improved answer quality

Key Technologies

• Azure Database for PostgreSQL: Database platform with AI extensions
• Microsoft Agent Framework: Framework for building AI agents with function tools
• Azure OpenAI: For embeddings and LLM chat completions
• PostgreSQL Extensions: Vector and Graph capabilities (DiskANN, AGE)
• Python: For agent development and database interaction

Project Structure

├── LICENSE                     # MIT License file
├── azure.yaml                  # Azure Developer CLI configuration
├── requirements.txt            # Python package dependencies
├── Code/
│   └── lab.ipynb              # Main Jupyter notebook tutorial
├── Dataset/
│   └── cases.csv              # Legal case dataset
├── Docs/
│   └── lab_manual.md          # Detailed lab instructions
│   └── images/                # Architecture diagrams and screenshots
├── Scripts/
│   ├── create_graph.sql       # SQL script for graph creation
│   ├── get_env.ps1            # Script to get environment variables
│   ├── get_user_token.ps1     # Script to get PostgreSQL access token
│   ├── initialize_dataset.sql # Database initialization script
│   ├── load_age.ps1           # Apache AGE installation script
│   └── write_env.ps1          # Environment variable setup script
├── infra/
│   ├── deployment_script.ps1  # Deployment script
│   ├── main.bicep             # Main Azure Bicep deployment template
│   ├── main.parameters.json   # Bicep parameters file
│   └── pg.bicep               # PostgreSQL-specific Bicep template
└── Archive/                   # Legacy scripts and configurations

Prerequisites

• Azure subscription with access to Azure OpenAI
• Visual Studio Code with the PostgreSQL extension
• Python environment with necessary libraries:
  • PostgreSQL connectivity (psycopg, psycopg-binary, psycopg-pool)
  • Modeling and validation (pydantic)
  • OpenAI and Agent Framework integration (openai, agent-framework)
  • Notebook compatibility (nest_asyncio, ipykernel)
  • Azure authentication (azure-identity)
  • Environment management (python-dotenv)

Lab Sections

1. Setup Azure PostgreSQL Database:

   • Database connection and configuration
   • Install the azure_ai extension
   • Configure Azure OpenAI connectivity

2. Using AI-driven features in PostgreSQL:

   • Pattern matching queries
   • Semantic vector search using embeddings
   • DiskANN indexing for fast vector similarity search

3. Building the Agent Application:

   • Setting up Microsoft Agent Framework
   • Creating database search AI function tools
   • Implementing semantic reranking
   • Adding external data sources (weather API)
   • Testing and improving the agent

Deploying to Azure

Follow these steps to deploy a PostgreSQL Flexible Server to Azure with the pgvector extension enabled:

1. Login to your Azure account:

   azd auth login

2. Create a new azd environment:

   azd env new

   Enter a name that will be used for the resource group. This will create a new folder in the .azure folder, and set it as the active environment for any calls to azd going forward.

3. Run this command to provision all the resources:

   azd provision

   This will create a new resource group, and create the PostgreSQL Flexible server and Azure OpenAI resources in that resource group. It will also create a .env file in the root of the project, which contains the connection information for the PostgreSQL server and Azure OpenAI.

Getting Started

1. Follow the instructions in Docs/lab_manual.md to setup your environment
2. Open Code/lab.ipynb in Visual Studio Code to follow the step-by-step guide
3. Run the PowerShell scripts in the Scripts/ folder to configure your environment and database

Additional Resources

• GraphRAG Solution for Azure Database for PostgreSQL
• Graph data in Azure Database for PostgreSQL
• PostgreSQL extension for Visual Studio Code
• Microsoft Agent Framework Documentation

License

This project is licensed under the MIT License - see the LICENSE file for details.