domus

Inspiration

Managing a home involves a lot of invisible work — planning, remembering, coordinating — that rarely gets acknowledged but takes real time and mental energy.
I wanted to explore what it would look like if more of that background work could be handled automatically, so people could spend less time managing logistics and more time on family, hobbies, and themselves.

Domus started from a simple question: what if running a home were treated as a coordination problem, not just a set of reminders?

What it does

Domus is a multi-agent home management system designed to reduce everyday decision overhead.
Instead of handling tasks in isolation, Domus coordinates specialized agents across events, inventory, and context. These agents operate over a shared home state and reason together to take action — whether that’s planning ahead for an event, surfacing the right reminder at the right moment, or translating inventory into decisions.

The focus is not on giving users more information, but on reducing the number of small decisions they need to make each day.

How I built it

The backend is implemented in Python and is responsible for agent orchestration, shared state management, and integrations with external services.
The frontend is built in React and focuses on clearly visualizing system decisions and agent outputs without overwhelming the user.

I designed Domus around an orchestration layer that allows domain-specific agents to operate independently while still coordinating through shared state. This made it easier to reason about system behavior and extend the architecture as new agents are added.

Challenges I ran into

I ran into API rate limits when coordinating multiple agents that rely on external services, which forced me to be intentional about batching, caching, and deciding when actions should actually be triggered.

Integrating with Blink IoT devices was another challenge. While it worked well enough to validate the idea of passive sensing, real-world conditions like lighting, angles, and device placement made it clear that input reliability is a hard problem. This reinforced the importance of decoupling sensing from reasoning so the system can evolve over time.

Accomplishments that I'm proud of

I’m proud that Domus is designed as a system, not a collection of features.
The current architecture cleanly supports multiple agents operating over shared state, and it’s structured in a way that can scale to additional domains without major rewrites.

I’m also proud of how clearly the demo communicates system behavior — focusing on reasoning and outcomes rather than just UI interactions.

What I learned

I learned that many everyday home problems aren’t about missing information, but about coordination and timing.
I also learned the importance of separating input quality from system intelligence — even imperfect data can be useful if the reasoning layer is well designed.

From a product perspective, I learned that people respond more strongly to systems that quietly remove work than to tools that ask for more attention.

What's next for Domus

Next, I want to expand Domus beyond the fridge agent by introducing additional agents and real-world user testing to validate where the system provides the most value.

I plan to support multiple users within the same household, explore expiration-aware reminders to reduce food waste, and experiment with using different models for different tasks to balance capability and cost.

I also want to support voice-based interaction and provide native integrations with mobile platforms, Google Home, and Alexa.

Longer term, I want to grow Domus into a broader home ecosystem — extending into areas like safety, security, and smart device coordination — and explore opportunities such as incubators to continue developing the system with real user feedback.

Built With

• gemini
• python
• react-native