Inspiration Living in Indonesia, I've seen firsthand how messy the agriculture supply chain can be, especially for commodities like shallots. A lot of data gets lost because logistics systems are too complex for everyday users. I wanted to build something where anyone could manage logistics simply by typing what they need in their local language, just like chatting with a friend.
What it does AgriSync acts as an AI logistics assistant. Instead of clicking through dozens of menus to update inventory or create a shipment, you just tell the Gemini agent what to do in plain text. The agent understands the intent, maps the data, and automatically triggers database operations to update the system in real-time.
How we built it I used Google's Agent Platform with the Gemini 2.5 Flash model as the brain.
For the database, I integrated MongoDB Atlas using Gemini's Tool Calling feature so the AI can directly interact with the database.
I built a simple React dashboard to prove that the database state updates immediately after the AI executes a command.
Finally, everything is deployed on Google Cloud Run using buildpacks for the Node.js environment.
Challenges we ran into The hardest part was getting the OpenAPI schema exactly right for Gemini's Tool Calling. The agent kept throwing '400 Invalid Argument' errors until I strictly defined the JSON structures and parameter types. Also, figuring out the deployment on Google Cloud Run and configuring the public access took some trial and error.
Accomplishments that we're proud of I'm really proud of making the tool-calling actually work flawlessly. Seeing the AI successfully understand an Indonesian command and automatically execute a CRUD operation into MongoDB Atlas for the first time felt amazing.
What we learned I learned a lot about defining strict JSON schemas for AI tools and how to connect an LLM to a real-world database instead of just using it as a simple chatbot. I also got much more comfortable navigating the Google Cloud ecosystem.
What's next for AgriSync AI Agent I want to integrate this agent into a WhatsApp or Telegram bot. This way, agricultural workers and truck drivers won't even need to open a web dashboard; they can just send a voice note or text message from the field to update the supply chain database.
Built With
- gemini-2.5-flash
- google-cloud-run
- mongodb-atlas
- node.js
Log in or sign up for Devpost to join the conversation.