HOMEVERSE — Think Before You Act

Inspiration

Smart homes are becoming more intelligent, but giving an AI control over real devices introduces an important challenge: what happens when an action creates an unexpected conflict?

Turning on multiple devices could exceed an energy limit. Adjusting a speaker could disturb someone sleeping. A seemingly simple command could create consequences the AI did not anticipate.

We built HOMEVERSE around one principle: an AI should think before it acts.

HOMEVERSE is a smart-home digital twin prototype that simulates actions, evaluates constraints, identifies conflicts, and supports safer decision-making before a simulated device action is applied.

What it does

HOMEVERSE creates a simulated smart-home environment containing six devices:

Living Room TV
Living Room Lights
Bedroom Lights
Air Conditioner
Bedroom Speaker
Hallway Smart Plug

Instead of treating every command as an instruction to execute immediately, HOMEVERSE follows a safety-oriented workflow:

Understand: Receive a requested smart-home action.
Simulate: Evaluate the proposed plan against the current home state.
Detect conflicts: Check constraints such as energy limits, bedroom noise limits, and hallway rules.
Replan: Identify a safer alternative when the proposed action conflicts with a constraint.
Request approval: Require explicit approval before applying a plan.
Verify: Inspect the resulting simulated device state after execution.

The goal is to make AI-driven automation more predictable, transparent, and controllable.

How we built it

The backend is built with Python, FastAPI, and Pydantic. FastAPI exposes endpoints for reading the home state, simulating plans, applying approved plans, and verifying execution results.

We also implemented a Model Context Protocol (MCP) server using Streamable HTTP. Its tools expose home-state retrieval, plan simulation, constraint checking, approved-plan application, and state verification to compatible MCP clients.

A shared home-engine layer manages the simulated devices and their constraints, keeping the core logic consistent across the API and MCP interfaces.

The project is designed around a clear separation between proposing an action and authorizing its execution. An AI-generated plan must not bypass the approval and validation workflow.

Challenges we faced

One of our main challenges was making the simulation, approval, and verification stages work together reliably. A proposed plan should not be treated as successful simply because the system generated it.

We addressed this by separating simulation from application and providing a dedicated verification step. We also tested the MCP tools and the rejection of unapproved plans.

What we learned

Building HOMEVERSE helped us explore API design, structured data validation, MCP tool integration, constraint-based planning, and safety-oriented automation.

We learned that useful AI automation is not only about generating the right action. It is also about checking whether the action is safe, asking for authorization, and verifying what happened afterward.

What's next

Our next steps include refining the dashboard, improving natural-language planning, expanding the simulated device environment, and exploring integrations with cloud AI services and real smart-home platforms.

HOMEVERSE is currently a simulated smart-home prototype, not a claim of production-ready control over physical devices. Our long-term vision is an AI-powered home that does not just act intelligently—it acts responsibly.

HOMEVERSE: Simulate first. Validate every action. Ask before executing. Verify the result.