Trader Network

Connect perishable stock items with buyers before it loses value, with agent assistance and human approval.

Comment

Inspiration

Perishable items get spoilt or have to be thrown away, mostly because traders can't find buyers nearby searching for exactly the stock they have on hand.

Trader Network is designed to connect that missing piece and timely connection between the buyer and the seller for a specific period of time before the produce gets thrown away. This platform grew from exploring a problem with informal market traders (including a fruit seller I usually buy fruits from), to a solution where humans and agents can interact seamlessly. I wanted to solve this issue using code AI agents without replacing the face-to-face transactions people already understand.

What it does

Trader Network helps sellers connect surplus or time-sensitive (about to expire/spoil) perishable produce with buyers.

  • Sellers are able to list their stocks, quantities, prices, and the available market window for the items.
  • Buyers can browse through available goods or even publish what they need.

Basically, both sides can prepare, review, and track offers.

What's better is that through the WebMCP, a browser agent can inspect the inventory, find its matching demand, and exxplain incoming offers, and prepare editable drafts. In essence, the agents draft the response; people can decide what gets approved

How we built it

I built the frontend with React, TypeScript, and CSS. Deployed on Cloudflare Pages, Pages functions handle the API, while Cloudflare D1 stores the pilot participants, inventory, requests, and offers.

For the judges, if you would like to test the live pilot, please use the code: "water"

The app exposes 9 agent tools through document.modelContext.registerTool() and the WebMCP callbacks share app state so that an agent's draft appears in the same workspace the person is using.

An anonymous demo sandbox lets the judges test both buyer and seller workflows without accessing the real and live data or changing the pilot records present in the database.

Challenges we ran into

One major challenge was differentiating browser support from agent access.

Another challenge was keeping the fictional data separate from live recorded data while making both easy enough to test and understand.

Accomplishments that we're proud of

I built this platform to support both ordinary forms and browser-agent assistance. This makes me proud because the working buyer-seller workflow allows consumers and sellers to interact in a unique way without the need to go to a supermarket or kiosk.

What we learned

I learned that an agent-friendly app does not necessarily need you to embed a chatbot. What it needs is useful features that an agent can invoke easily.

What's next for Trader Network

The next step is a small trial involving independent and consenting buyers and sellers at large. Measurements need to be taken to determine whether or not people find suitable matches, successfully contact each other, and complete pickups. Scaling to a global level is also part of the goals for Trader Network, where traders and buyers can use the platform from different parts of the world.

Share this project:

Updates

Submission history