🚀 SmartSupply AI: Multi-Agent Supply Chain Intelligence using Google ADK
Built for the ADK Hackathon using Google’s Agent Development Kit (Python).
🏷️ #adkhackathon #multiaagents #GoogleCloud
🌟 Inspiration
The complexity of global supply chains — with their diverse geographies, regulations, and dependencies — inspired us to build SmartSupply AI, a multi-agent system designed to automate logistics intelligence and optimization. Our goal was to make supply chain decision-making more data-driven, resilient, and real-time.
🤖 What It Does
SmartSupply AI is a multi-agent system powered by LLMs, purpose-built to streamline and simulate real-world supply chain logistics.
It allows users to:
- 🧭 Query global logistics routes (shipping, customs, trucking, warehousing).
- ⚠️ Score risks based on delays, compliance, bottlenecks.
- 🔁 Recommend supply chain optimizations for cost, time, and risk.
- 📝 Collaboratively generate structured insights from agent orchestration.
✅ Supports CLI and Web UI
🌍 Live Demo: SmartSupply AI Web App
🏗️ How We Built It
We used the Python version of Google’s Agent Development Kit (ADK) to orchestrate multiple agents:
- Fetch Agent → Pulls contextual data based on prompts.
- Risk Agent → Uses domain heuristics to assess route/product risks.
- Optimization Agent → Suggests improvements to reduce delays and costs.
- Report Agent → Summarizes results in readable formats.
We deployed the system on Render and provided a CLI and web interface.
⚙️ Architecture
🧠 Sample Prompts for Real-World Testing
Try these to experience the full power of agent coordination:
- 📦 Smartphones from Shenzhen to Chennai
- 🖥 Monitors from Taiwan to Noida
- 💊 Pharma APIs (Cold Chain) from Germany to Bhiwandi
- 👗 Textiles from Dhaka to Kolkata (via Petrapole)
- 🥤 Packaged Beverages from Pune to Ahmedabad
Copy/paste these into the app to test real-world multi-agent analysis.
🧗 Challenges We Faced
- Understanding agent chaining and memory persistence with ADK.
- Managing external credentials (Google API keys, service accounts).
- Handling cold-chain pharma routing complexity.
- Deploying the system with minimal latency on Render.
🏆 Accomplishments
- Successfully deployed a real, working multi-agent logistics intelligence platform.
- Created reusable components using ADK’s SequentialAgent pipeline.
- Integrated LLM tools and domain-specific logic without fine-tuning.
📘 What We Learned
- Deep dive into the ADK’s modular agent orchestration capabilities.
- Efficient multi-agent task delegation and result consolidation.
- Leveraging language models for structured B2B tasks, not just chat.
🔮 What's Next
- Add PDF export (via GCS Artifacts).
- Plug in real logistics APIs (e.g., Freightos, Portcast).
- Enable visual dashboards (charts, maps).
- Extend to include carbon emissions modeling.
🛠️ Built With
📦 GitHub Repository
Clone and test using:
bash
git clone https://github.com/jayaram9196/DemoAgent.git
cd DemoAgent
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
adk web # to start the web interface
Top comments (0)