Yuyabre

An AI-powered "smart roommate" built in 24 hours at the Prosus AI Hackathon 2025. It handles natural-language commands, tracks pantry inventory, coordinates group grocery orders, and logs shared expenses to Splitwise, all through a chat interface backed by a tool-calling LLM agent.

Yuyabre is a prototype AI roommate for shared flats, built with a friend during the Prosus AI Hackathon 2025 in 24 hours. Users interact with a chat interface to manage household life through natural language.

1. LLM agent

   • POST /agent/command for synchronous replies; WS /agent/command/stream for streamed tokens in the React UI

   • Tool-calling architecture: the model can invoke ~15 backend tools covering inventory, orders, housemate coordination, Splitwise, Discord, and WhatsApp

   • Tools include inventory snapshots, low-stock checks, grocery order placement, group-order coordination with housemate responses, and expense logging

2. Household features

   • Per-user and shared inventory with categories, thresholds, expiration dates, and restock alerts

   • Grocery ordering bridge (Thuisbezorgd-oriented, mocked for demo) with order history and simulated delivery ETA

   • Splitwise OAuth integration for automatic shared-expense logging after group orders

   • Discord webhook for group-order notifications and yes/no response polling; Twilio WhatsApp as fallback

3. Tech stack

   • FastAPI backend with Beanie/MongoDB; async LRU cache with write-through invalidation

   • React + Vite frontend with TanStack Query, markdown rendering, and PWA support

   • Docker Compose for local deployment (API + MongoDB + Mongo Express)

   • Interactive Rich CLI for agent demos without a browser