OptiChain AI: Transforming Supply Chains with AI

Inspiration

The idea for OptiChain AI was born from the global supply chain crises of recent years—delayed shipments, empty shelves, and soaring costs. As an AI enthusiast, I saw an opportunity to harness Google’s ADK and Gemini’s advanced capabilities to create an autonomous platform that optimizes supply chains in real-time, reducing waste and promoting sustainability. The challenge of solving a billion-dollar problem with cutting-edge tech inspired me to build a solution for industries like retail and manufacturing.

What it does

OptiChain AI is a multi-agent platform that automates end-to-end supply chain processes. It features:

Demand Forecasting: Predicts sales using historical data (BigQuery) and real-time market trends (Firecrawl). Inventory Management: Monitors stock levels with real-time API queries and JSON reports. Logistics Optimization: Suggests fuel-efficient routes using Google Maps and simulated traffic analysis. Supplier Communication: Automates negotiations with long-context natural language processing. Orchestration: Coordinates agents to adapt to disruptions like weather delays. A React-based UI visualizes insights, making it accessible for businesses to optimize operations.

How we built it

We developed OptiChain AI with a modern, scalable tech stack: Backend: FastAPI powers a Google ADK multi-agent system, with agents leveraging Gemini via Vertex AI for structured outputs, RAG, and function calling. Python Faker generated mock data for sales, inventory, and suppliers. BigQuery and AlloyDB handled historical and real-time data. Frontend: A React app with Material-UI delivers a professional UI, featuring Chart.js for forecasts, Google Maps for routes, and a chat interface for supplier interactions. Deployment: The backend runs on Cloud Run, and the frontend is hosted on Firebase Hosting. GenAI Features: We implemented JSON reports, document parsing, simulated video analysis, RAG with Firecrawl, long context windows, and context caching, all tested in Replit.

Challenges we ran into

Agent Synchronization: Coordinating agents to handle disruptions required complex orchestrator logic. Gemini Multimodal Setup: Simulating video understanding for traffic analysis with mock data was tricky without real feeds. Data Realism: Crafting realistic mock data with Faker to reflect supply chain dynamics took careful tuning. Latency: Balancing RAG and context caching for real-time performance was computationally intensive. UI Design: Achieving an enterprise-grade, responsive UI demanded multiple iterations with Material-UI.

Accomplishments that we're proud of

• Built a scalable multi-agent system that seamlessly integrates Gemini’s advanced features like RAG and document understanding.
• Created a professional, intuitive UI that visualizes complex supply chain data effectively.
• Successfully simulated real-world disruptions, showcasing the platform’s adaptability.
• Leveraged Google Cloud tools (BigQuery, Vertex AI, AlloyDB) for a robust, cloud-native solution.
• Prepared a compelling demo video highlighting real-time optimization, ready for #adkhackathon.

What we learned

• Mastered Google ADK for multi-agent orchestration and Gemini for multimodal AI tasks.
• Gained expertise in Google Cloud services for data storage, AI deployment, and scalability.
• Learned to design realistic mock datasets with Python Faker for testing.
• Improved skills in building modern web apps with React, Material-UI, and FastAPI.
• Developed problem-solving strategies for optimizing latency and agent coordination.

What's next for OptiChain AI

• Add new agents, like a sustainability tracker to monitor carbon footprints.
• Integrate real-time data feeds (e.g., live traffic, supplier APIs) for production use.
• Enhance UI with advanced analytics dashboards for deeper insights.
• Contribute a sample agent to the ADK open-source repository.
• Publish a blog post with #adkhackathon to share our journey and inspire others.

Built With

• alloydb
• bigquery
• chart.js
• cloud-run
• fastapi
• firebase-hosting
• firecrawl
• firecrawl-api
• gemini-ai
• google-adk
• google-cloud
• google-maps
• googlemaps/react-wrapper
• html/css
• javascript/typescript
• material-ui
• python
• python-faker
• react
• replit
• simulated-warehouse-apis
• vertex-ai
• vertex-ai-api