Multi-Agent System for Smart Kitchen Renovation

A collaborative AI-powered multi-agent system built using Google Vertex AI, ADK, AlloyDB, MCP Toolbox, and Gemini 2.5, designed to automate the process of kitchen renovation through intelligent agents.

🧠 Built as part of Code Vipassana Season 10 organized by GDG Cloud Kochi

Link to my Code Vipassana Certificate

---

🧩 Project Overview

This project simulates a smart kitchen renovation workflow with three autonomous agents, coordinated by a Root Agent:

1. 📄 Renovation Proposal Agent

   • Generates a kitchen renovation proposal document based on the user's specifications.
   • Uploads the generated PDF to a Google Cloud Storage bucket.

2. ✅ Permits & Compliance Check Agent

   • Validates renovation plans against local regulations.
   • Creates checklists and drafts official emails for authorities.

3. 📦 Order Status Check Agent

   • Retrieves order status from an AlloyDB procurement database.
   • Built using MCP Toolbox and Cloud Run Functions.

---

⚙️ Technologies Used

S.No	Technology	Used For
1	Vertex AI & Gemini 2.5	Powers the chat-based multi-agent system for intelligent response generation and document creation.
2	Google ADK (Agent Developer Kit)	Builds and orchestrates multi-agent workflows with sub-agent delegation and conversation management.
3	Google Cloud Run	Deploys the root agent and backend services in a scalable, serverless environment.
4	MCP Toolbox for Databases	Connects to AlloyDB and enables secure, tool-based SQL querying for material order status.
5	AlloyDB	Acts as the backend database for procurement and delivery tracking (material_order_status table).
6	Google Cloud Functions	Provides serverless APIs to fetch data from AlloyDB (used by the ordering agent).
7	Google Cloud Storage	Stores the generated proposal PDFs uploaded by the Proposal Agent.
8	LangChain Toolbox	Supports natural language tools and actions for enhanced agent capabilities.
9	Python	Python handles agent orchestration and logic.
10	Java	Java is used in the Cloud Function for database access.

---

🚀 How to Run

CLI

adk run .

Web UI (ADK Provisioned)

adk web

Ensure your environment variables and credentials are configured (use a .env file with necessary keys and bucket details).

---

🧠 Agent Architecture

Root Agent
│
├── Renovation Proposal Agent (PDF creation + GCS upload)
├── Permits and Compliance Agent (Permits checklist + email)
└── Order Status Agent (AlloyDB + MCP Toolbox integration)

---

🖼️ Result Screenshots

1. Renovation proposal agent output

2. Proposal PDF uploaded in Cloud Storage bucket

3. Root agent web UI output

4. Procurement order status agent output

5. AlloyDB table for procurement material order status

---

📂 Project Structure

.
├── agent.py                   # Root agent + sub-agents
├── mcp-toolbox/              
│   ├── mcp_agent.py          # MCP agent logic
│   └── tools.yaml            # AlloyDB connection config
├── Cloud_Run_Function/
│   ├── ProposalOrdersTool.java # Cloud Function for order status
│   └── pom.xml
├── database_script.sql       # DDL + sample inserts for AlloyDB
├── requirements.txt
└── README.md

---

🧑‍💻 Author

Built with ❤️ by Fathima Zulaikha during Code Vipassana Season 10
👉 Organized by GDG Cloud Kochi

---