AgriFlow

Inspiration

From the seven years of hands-on experience working with rural smallholders in Uganda, I witnessed a recurring systemic failure: the Production-Market Mismatch. Farmers often invest their limited resources into crops without real-time market data, leading to "off-market" produce that fails to sell. This is compounded by lack of technical production support and fragmented supply chains. AgriFlow was born from the need to turn these data-poor environments into data-driven ecosystems.

What it does

AgriFlow is an end-to-end AI platform designed to empower farmers from seed to sale. It acts as an intelligent bridge between production and procurement:

For Farmers: Provides a Voice User Interface (VUI) and Multi-lingual AI Agent support, ensuring accessibility for users with varying literacy levels. The agent offers personalized agronomic advice and real-time market price signals.

For Buyers: Offers a robust dashboard for Aggregation Management and Traceability, providing quality assurance and logistics tracking that was previously impossible in fragmented rural markets.

How we built it

The application was developed using an Agentic "Vibe Coding" workflow within Google AI Studio, leveraging the Gemini 3 Pro model.

AI Orchestration: Gemini 3’s advanced reasoning handles complex multimodal queries (voice and text) to provide localized agricultural insights.

Agentic Reasoning: The Gemini 3 "Deep Think" capability acts as an orchestrator, managing complex logistics and "aggregation logic" to group small yields into marketable lots for buyers.

Backend & Data: Powered by Node.js and Firebase Firestore for real-time data synchronization across the supply chain.

Frontend: Built with TypeScript, ensuring a type-safe and responsive interface that facilitates seamless interaction between the AI agent and the user.

Challenges we ran into

The primary hurdle was managing high-compute resource requirements within limited GCP credits. Optimizing the token usage of the Gemini 3 agent while maintaining its "thinking" depth required rigorous prompt engineering and architectural trade-offs to ensure the prototype remained functional and cost-effective.

Accomplishments that we're proud of

We successfully moved from concept to a fully functional prototype in record time. The integration of the Voice UI in local Ugandan dialects via Gemini’s multilingual capabilities proved that advanced AI can be made accessible to the most remote users.

What we learned

This project proved that Rapid Prototyping and "Vibe Coding" are not just trends—they are transformative for solo engineers. By using Gemini 3 as a co-architect, I could focus on high-level system design and data logic while the AI handled the boilerplate, significantly accelerating the development lifecycle.

What's next for AgriFlow

The immediate goal is field validation. We are moving into a pilot phase to test the app with real users in rural Uganda, gathering data to refine the AI Agent’s reasoning and scaling our infrastructure to support regional aggregation hubs.

Built With

• antigravity
• cloudrun
• gcp
• gemini
• google-cloud
• googleaistudio
• node.js
• typescript