Problem
Canada’s local food market is still difficult to access for both farmers and restaurants. According to Statistics Canada, only 7.8% of Canadian farm operations sold food locally, showing that local sales channels remain limited nationwide. The Ontario Federation of Agriculture found that among farmers using off-farm market channels, only 25% sold direct to restaurants, with farmers citing advertising, marketing and time constraints as key barriers. On the demand side, another study found that 93% of restaurants said they would use more local food if it were easier to access. Together, these findings point to the same gap: restaurants want local supply, but farmers face real friction reaching them directly. The result is a fragmented local food marketplace that makes local procurement harder than it should be for both sides.
Our Solution
Farmesh bridges that gap by giving restaurants and farmers a direct, intelligent marketplace. Restaurants post what they need, farmers post what they have, and Backboard-powered agent orchestration generates tailored matches between demand and supply. By storing context across sessions and sharing information across agents, the platform continuously improves recommendations based on preferences and feedback. With separate, personalized experiences for vendors and customers, Farm Match makes local sourcing simpler, faster, and more practical for both sides.
How We Built It
- Frontend: Next.js + Tailwind CSS
- Backend: Node.js
- User Auth: Supabase
- Database: Supabase
- Agent Orchestration: Backboard
- Agent Models: Gemini 2.5-Flash
- IDE: Antigravity
How We Used Antigravity
We utilized the built in Claude Opus and Sonnet models for complex coding tasks, and Gemini 3.1 Pro for planning agent tasks and logic. The Antigravity IDE allowed us to parse through Backboard documentation and use cases autonomously to provide walkthroughs for leveraging the API while simultaneously implementing it. We also had it generate design documents for each individual agent which we exported to share and tweak to our liking. Our testing process also utilized the Antigravity IDE, as we could generate detailed test scenarios and edge cases considerations using Gemini with each iteration in our code.
Challenges We Ran Into
Our initial biggest challenge was finding an idea that targeted a niche, underserved community while still being generalizable to the whole Canadian population. We also faced challenges trying to not compromise code efficiency and performance with crafting a complex, interactive agentic infrastructure that would mesh well with our database structure. Specifically, we struggled to implement the normalization agent to ensure uniform unit conversion across a vast array of inputs while also organizing and storing this information meaningfully and consistently.
What We Learned
We spent a lot of time learning how to structure our database to support scalable storage and retrieve and fill in parsed information from each agent. Moreover, we had to learn prompt design/engineering skills to fine tune the scope of our AI agents. Decision-making and system design was also a large area of growth for us, as we often had to adapt our approach and quickly reprioritize and restructure our project collaboratively.
Future Improvements
In the future, we hope to introduce precise location-based matching to create more accurate vendor-customer pairings while also helping optimize transportation and delivery costs. We also plan to build a third administrative interface that would allow platform administrators to monitor market activity, track supply and demand trends, and generate broader economic insights from platform data. In addition, we want to develop an email-drafting agent that uses match details, vendor information, and customer preferences to automatically generate personalized outreach and follow-up messages.
Revenue Model
To encourage adoption, Farmesh would initially be offered free of charge to farmers and restaurants. This lowers the barrier to entry and helps us build both sides of the marketplace. Over time, our main revenue stream would come from delivery and logistics services. Once users begin transacting through the platform, Farmesh could coordinate transportation and fulfillment, earning revenue through delivery fees, logistics margins, or service charges tied to successful shipments.
Log in or sign up for Devpost to join the conversation.