• 

  SmartSupplyAI Architecture Diagram

SmartSupply AI

AI-powered multi-agent supply chain advisor system built using Google's Agent Development Kit (ADK)

🚀 Inspiration

Global supply chains are under constant pressure from geopolitical issues, weather disruptions, port congestion, and rising costs. We wanted to build an AI-driven tool that can intelligently analyze supply chain vulnerabilities and provide actionable insights—just like a real team of logistics analysts and planners would.

💡 What It Does

SmartSupply AI is a modular, LLM-powered multi-agent system that:

• Collects relevant logistics and market data for a given region and product
• Assesses supply chain risk based on cost, delays, transportation issues, and regulations
• Provides optimization suggestions such as route diversification, demand forecasting, and buffer strategies
• Compiles all outputs into a clean, structured, human-readable report

🛠 How We Built It

Built with:

• Google ADK (Agent Development Kit) for managing intelligent agents
• LiteLlm model interface using Gemini 1.5 Flash
• A SequentialAgent pipeline of four agents:
  1. fetch_logistics_data
  2. analyze_risks
  3. optimize_supply_chain
  4. generate_supply_chain_report
• A tool function: get_supply_chain_risk_score() for contextual risk scoring

Each agent is isolated, with clearly defined prompts and tasks to maintain modularity and context integrity.

⚠️ Challenges We Ran Into

• Initial setup and authentication with ADK and API keys
• Maintaining shared state across multi-step agents
• Designing agents with minimal overlap but coherent communication
• Generating useful outputs from simulated data (due to no live API integration)

🏆 Accomplishments That We're Proud Of

• Designed and implemented a realistic simulation of enterprise-level supply chain workflows
• Used ADK to build a fully modular and extensible agent system
• Achieved smooth context passing and robust instruction chaining across agents
• Delivered meaningful insights from simple structured inputs

📚 What We Learned

• How to chain multiple LLM agents in a production-like flow
• Best practices with Google’s ADK framework and tool registry
• How to model risk heuristics, transportation logic, and inventory assumptions in prompts
• The value of clear agent instruction design in long-chain reasoning

🔮 What’s Next for SmartSupply AI

• 🌐 Integrate with real-time logistics APIs (e.g., port data, freight cost, weather conditions)
• 🖥 Build a web UI for easier user interaction and result display
• 🏭 Expand templates for industry-specific use cases like pharma, electronics, or apparel
• 🌍 Enable multi-language agent flows for global operations
• 🧠 Explore non-linear agent collaboration (parallelism, feedback loops, retries)

Built With

• dotenv
• gemini-ai
• github
• google-adk
• google-cloud
• litellm
• python
• vscode