Eat With Cat

The AI agent (built using CDP's AgentKit) generates Doordash food delivery recommendations for a user, given their dish preference and location. Upon user's agreement, the agent places an order on Doordash via its API from the recommended restaurant, and pays for the order on chain via CDP. The recommendation engine saves users the hassle of browsing through endless restaurants on delivery apps, as well as eliminates the need for a Doordash account altogether.

This project is built as a Telegram chat bot and mini app, so we heavily used Telegram's API to send and receive messages via the bot. After a user runs /start command which kicks off the interaction with the bot, we continuously poll the conversation for updates (new texts) from the user, and react to them accordingly. The bot is utilizing Telegram web hooks to get updates, which we implemented using a ngrok endpoint. The bot logic is built in Python, while the mini app front end uses JavaScript.

As we guide the user through our flow, we get their Coinbase Wallet information, as well as their food preference, address, etc. Then, the AI agent (built using CDP's AgentKit) generates restaurant recommendations for the user using this information, and we send this as a message in the chat. The user is able to decline the restaurant, which will prompt the AI to generate a new one. On the other hand, the user can approve the food delivery from a restaurant, at which point the AI agent will use Doordash API to place this order!

Coinbase Wallet and CDP's AgentKit have brought out project to life, as they enable the bot to generate recommendations, accept on chain payments, and place orders.

We believe that the coolest part of our project is how independent our AI agent is! From paying for food on chain, to actually placing orders on Doordash, Eat With Cat is hacky and awesome all around!