🚀 Google ADK Python Agentic App: Autonomous Procurement System

A Next-Gen Agentic AI System built with Google ADK (Agent Development Kit) and Python. This ADK Python project demonstrates how to build autonomous agents that handle complex enterprise workflows...

This project demonstrates the future of enterprise software: Autonomous Agents that don't just chat, but work. It implements an intelligent procurement assistant capable of handling complex purchase requests, negotiating with suppliers, and managing approval workflows—all with a seamless Human-in-the-Loop (HITL) interface.

---

✨ Key Features

🧠 True Agentic Reasoning

Powered by Google's Agent Development Kit (ADK) and Vertex AI (Gemini Pro), this agent goes beyond simple RAG.

• Multi-Step Planning: Breaks down complex requests (e.g., "I need 50 laptops for the new engineering team") into actionable steps.
• State Management: Maintains context across long-running conversations and workflows.
• Policy Compliance: Automatically checks requests against internal procurement policies.

🤝 Human-in-the-Loop (HITL) with AG-UI

Automation shouldn't be a black box. The AG-UI frontend provides a rich, streaming interface for collaboration.

• Interactive Cards: The agent streams structured UI components (approval forms, product comparisons) directly into the chat.
• Approval Workflows: Critical actions pause for user confirmation, ensuring control over budget and compliance.

🔌 Extensible MCP Ecosystem

Built on the Model Context Protocol (MCP), the system is modular and easily extensible.

• MongoDB MCP: Persistent memory for user profiles, purchase history, and supplier data.
• PDF Generation MCP: Dynamically generates professional Purchase Orders and reports.
• Search Tools: Integrated web search for real-time supplier discovery.

---

📸 Visual Walkthrough

1. Intelligent Need Analysis

The agent understands vague requests and asks clarifying questions to build a complete purchase requisition.

2. Smart Product Search & Comparison

It searches catalogs (and the web) to find the best products, presenting them in a clear comparison view.

3. Human-in-the-Loop Approval

Before finalizing, the agent presents a summary for approval. You can edit details or confirm with a single click.

4. Automated PO Generation

Once approved, the agent generates a formal Purchase Order PDF and handles the "paperwork" automatically.

---

🏗️ Architecture

The system consists of three main pillars:

1. Frontend (app/): A modern Angular application using AG-UI for streaming agent responses.
2. Procurement Agent (procurement/): The core reasoning engine built with Python and Google ADK.
3. MCP Servers (mcp_server/): Specialized microservices for Data (MongoDB) and Documents (PDF).

---

🚀 Quick Start

Prerequisites

• Node.js (v18+)
• Python (v3.11+)
• Yarn (Installed automatically if missing)
• Google Cloud Project (Vertex AI API enabled)

Setup & Run

1. Configure Environment: Copy the .env.example files to .env in procurement/, mcp_server/mongodb/, and mcp_server/pdf-generation/, and fill in your API keys.

2. One-Click Setup:

   ./setup.sh

   Installs all dependencies (Python & Node) and sets up the environment.

3. Run the System:

   ./run.sh

   Starts the Agent, MCP Servers, and Frontend.

Access the app at http://localhost:4200!

---

☁️ Deployment

Ready for production?

• Google Cloud Run: ./deploy_cloudrun.sh (Deploys as containerized services)
• Vertex AI: ./deploy.sh (Deploys reasoning engine to Vertex AI)