Family Hub: Tentacle OS

Inspiration

Modern families are overwhelmed by "app fatigue"—juggling WhatsApp for chat, Splitwise for expenses, Google Photos for memories, and various notes apps for lists. We wanted to build Family Hub to consolidate these fragmented experiences into a single, private "operating system" for the home, where an AI doesn't just chat, but actually manages the household data.

What it does

Family Hub is a premium household command center. It provides:

Intelligent Chat: A real-time messaging interface where /ai commands trigger a multi-agent system.
Financial Management: An AI Accountant that queries transaction history and provides budget summaries.
Automated Archiving: A system that automatically tags and organizes uploaded files and photos into searchable albums.
Logistics & Planning: Specialized tools for travel itineraries and family calendars.
Security: Enterprise-grade Auth0 integration ensures family data stays within the family. ## How we built it We utilized a high-performance, modern stack:
Frontend: Next.js 16 (App Router) with Tailwind CSS 4 for a sleek, premium UI.
Backend: FastAPI (Python) handling heavy lifting, WebSockets for real-time updates, and SQLAlchemy for database management.
AI Engine: Google Gemini 2.5 Flash, implemented with a router-provider pattern to delegate tasks to specialist agents.
Authentication: Auth0 via Next.js API proxies to protect backend routes with JWT injection. ## Challenges we ran into
Stateful Agent Routing: Getting the MainAgent to accurately distinguish between general chat and specialized financial queries required fine-tuning the prompt engineering and context handling.
BFF Pattern: Synchronizing the Next.js frontend proxies with the FastAPI backend to ensure seamless - JWT passing without adding latency to the user experience.
Real-time Synchronization: Managing WebSocket connections across different family channels to ensure messages and AI responses appear instantly for all members.

Accomplishments that we're proud of

Specialist Agent Logic: Creating an "Archivist" that doesn't just store files, but actively understands them by generating metadata tags.

Unified Schema: Designing a database that elegantly links chat messages to photo albums and financial transactions.

Premium UX: Achieving a "pro-sumer" feel with smooth transitions and a clean, minimal interface that feels like a high-end productivity tool.

What we learned

We deepened our understanding of Multi-Agent Systems (MAS)—specifically how to manage "handoffs" between different LLM instances. We also learned the importance of the BFF pattern in protecting internal APIs while maintaining a fast, reactive frontend.