Skip to content

00_Introduction.md

Latest commit

 

History

History
102 lines (72 loc) · 5.55 KB

00_Introduction.md

File metadata and controls

102 lines (72 loc) · 5.55 KB

Introduction

Note

This is a 90-minute workshop that gives you hands-on experience building a Compliance RAG Agent using Microsoft AI Foundry, the Foundry Toolkit for VS Code, and GitHub Copilot. You will go from zero to a fully deployed compliance assistant with a web UI.

Learning Objectives

By the end of this workshop, you will be able to:

  • Provision Azure resources (AI Foundry, Blob Storage, Azure AI Search) required for a RAG agent.
  • Upload and index compliance documents into Azure AI Search from Azure Blob Storage.
  • Design and prototype an agent in Foundry Toolkit for VS Code Agent Builder with Azure AI Search as a grounding tool.
  • Export agent code and debug real-world runtime errors using GitHub Copilot.
  • Extend the agent with interactive input and a professional web UI using GitHub Copilot Agent mode.
  • Containerize the application for deployment using Docker (Optional).
  • Evaluate and refine agent responses through structured test scenarios.

Architecture Overview

flowchart LR
    U[" User Query"] -->|"1. Submit"| UI[" Streamlit Web UI\napp_ui.py"]
    UI -->|"2. Query"| A[" Azure AI Foundry\nAgent (GPT-4o)"]
    A -->|"3. Search"| S[" Azure AI Search\nRAG Retrieval"]
    S -->|"4. Retrieve"| B[" Blob Storage\nKB Documents"]
    B -.->|"5. Chunks"| S
    S -.->|"6. Context"| A
    A -.->|"7. Report"| UI
    UI -.->|"8. Display"| U
Loading

Azure = Brain (Models + Search) · Foundry Toolkit for VS Code = Agent Designer · GitHub Copilot = Builder + Debugger + Extender

Business Scenario

The Challenge

Compliance and risk teams in fintech, banking, and manufacturing organizations must evaluate regulatory risks across multiple jurisdictions before onboarding vendors, handling cross-border data transfers, or signing contracts. Today, this process involves manually searching dozens of regulatory documents, cross-referencing multiple policies, and writing compliance reports from scratch — leading to slow turnaround and inconsistent risk assessments.

The Solution: Compliance Compass

You will build Compliance Compass, an intelligent compliance RAG agent that:

  1. Accepts natural language queries — Users describe real-world compliance scenarios (e.g., "We are onboarding an AI vendor in China that will process customer payment data.")
  2. Retrieves relevant policies — The agent searches a curated Knowledge Base of 12 compliance documents via Azure AI Search.
  3. Analyzes risks — The agent reasons over retrieved policies to identify conflicts, obligations, and risks.
  4. Generates structured reports — The output includes risk scores, referenced policies, immediate actions, and escalation options.

Example Interaction

User Prompt:

"Assessing risks for a new AI analytics vendor based in Singapore that will process our customer transaction data. Are there any RBI localization concerns?"

Agent Report (summarized):

  • Risk Score: 7/10 (High)
  • Key Findings: RBI mandates payment data storage in India; processing abroad allowed but data must return within 24 hours.
  • Recommended Actions: Add data residency clause, conduct Transfer Impact Assessment, perform vendor cybersecurity audit.
  • References: RBI Data Localization 2018 Guidelines, GDPR Article 44 Transfers.

Workshop Flow

Tip

Complete the modules in order — each builds on the previous one.

Module Title What You Will Do Duration
I Prerequisites and Environment Setup Install tools, authenticate with Azure, configure VS Code extensions 10 min
II Provisioning Azure Resources Create AI Foundry hub, deploy models, set up Blob Storage and AI Search 15 min
III Designing the Agent in Foundry Toolkit for VS Code Build the compliance agent visually with instructions and Azure AI Search tool 10 min
IV Exporting Code and Debugging with Copilot Export to Python, encounter a real Azure SDK error, debug with Copilot 15 min
V Adding Interactive Input Transform hardcoded queries into an interactive CLI with Copilot 10 min
VI Building a Professional Web UI Generate a complete Streamlit chat UI using Copilot Agent mode 15 min
VII Containerization and Testing (Optional) Dockerize the app with Docker (optional), run quality tests, refine agent instructions 15 min

Total Duration: ~90 minutes

How the Three Tools Work Together

Technology Role in Workshop
Microsoft AI Foundry The brain — hosts GPT-4o (reasoning) and embedding model; manages agent deployment
Foundry Toolkit for VS Code The designer — visual builder to configure agents, instructions, and tools without writing code
GitHub Copilot (Agent Mode) The builder — debugs errors, generates code extensions, creates UI, writes Dockerfile
Azure AI Search The memory — indexes and retrieves relevant compliance policies for RAG grounding
Azure Blob Storage The storage — stores the 12 compliance Knowledge Base documents

Note

Instructor Note: Each module contains step-by-step instructions. Participants copy prompts and let GitHub Copilot generate the output — minimal manual coding is required.

Click Next to proceed to Module I: Prerequisites and Environment Setup.