BwanaShamba

Inspiration

In January 2026, I secured a small piece of land in Malivundo, Pwani, Tanzania, to begin a journey into agriculture. What should have been an exciting step quickly became overwhelming. I needed expert guidance to grow tomatoes and onions successfully, but agronomists are scarce, and hiring one full time is too expensive for a beginner. In February, I got an idea, what if an AI agent could fill this gap? Instead of relying on constant human supervision, a digital agronomist could provide guidance on tomato and onion farming while also automating tasks such as starting and stopping irrigation. This could reduce the cost of human agronomist intervention significantly. While researching how to build such a system, I discovered the Gemini Live Agent Challenge. That moment turned my personal project into a mission. Participating in this challenge is pushing me to build a solution that doesn't just solve my own problem, but has the potential to empower thousands of smallholder farmers who remain tied to traditional methods simply because professional expertise is currently inaccessible.

What it does

I have built the Agent to be my full time agronomist and a full time casual worker. So the Agent will provide all technical support for growing tomatoes and onions in Malivundo Pwani, and will also perform basic tasks such as irrigation, which will be automated based on time and weather conditions.

How we built it

With minimal understanding and no code experience, I used AI, mostly Gemini, (and other models when i need to compare notes), to learn on how AI agents work and vibe code using tools like Gemini AI studio to create my first draft of the Agent and move to Replit for further fine tuning and more help on deploying since Replit uses Google Cloud for its own deployment. After development, I pushed the code to GitHub and used Gemini to guide me in setting up the environment in Google Cloud Run and mapping the app to my recently procured domain.

Challenges we ran into

The biggest challenge is deploying. Spent a lot of time deploying, since there were quite a few things that Replit built that didn't work well when sent to Google Cloud Run. But again, with the use of Gemini in the cloud for support, the main (Pro) model for complex code problems, and Replit itself I was able to resolve these.

Accomplishments that we're proud of

Building this Agent. While the Gemini Live Agent Challenge might have sped up the process of building, this app is the most ambitious and useful project I have ever tried. It will definitely help me in my agriculture journey, and it has the potential to support and transform millions of Tanzanians facing challenges in Agriculture.

What we learned

Google Agent Development Kit has been key for this app to function so smoothly and with such precision. It was clear, building with and without Google ADK, how the application improved.

What's next for BwanaShamba

I will continue to improve on the application, engage with seed development firms in Tanzania and other parts of East Africa to get datasheets for their crops, farmers to get real life results of the crops, agronomists, and the Agriculture institutes to get verified information. This will provide cleaner data and prevent the agent from hallucinating if it fails to gather enough data in Tanzania to generate its response. I will also work to connect the application with the physical world to handle tasks such as an automated irrigation system.

Built With

• adk
• bcrypt
• cloudrun
• cloudsql
• css
• express.js
• gemini
• open-meteo-forecast
• postgresql
• python
• react
• rest
• sse
• tailwind
• vite