Orion Home AI

The Autonomous Representative for Home Maintenance

I built Orion because home maintenance is fundamentally a "language gap" problem. I wanted to create more than an app; I wanted to create an Autonomous Representative—a digital concierge that speaks the technical language of pros but works exclusively for the homeowner.

What it does

Orion is an Agentic AI Marketplace that automates the entire lifecycle of a home repair:

• Vision Diagnostics: Users upload a photo or video. Using Gemini 3 multimodal vision, Orion identifies the specific failure point, assigns a severity grade (1-4), and generates a "Pro-Brief"—a technical payload for the contractor.
• Autonomous Negotiator: This is the project's crown jewel. Once a problem is found, Orion acts as your agent. It uses Gemini 3 Function Calling to interview local specialists, transmitting the technical brief and negotiating arrival times and prices based on the user's budget.
• Capital Ledger: A high-tech system manages project funds in escrow, ensuring specialists are only paid once the AI verifies the structural integrity of the fix.
• Neural Inbox: A persistent, real-time chat architecture that syncs AI-assisted expert advice with a centralized message history.

How I built it

Orion is powered by the Gemini 3 Pro and Flash models via the Google GenAI SDK:

• The Brain (Gemini 3 Pro): I utilized the Pro model for high-reasoning tasks like failure analysis and the complex logic required for the Contractor Negotiator.
• Agentic Logic (Function Calling): I implemented a recursive tool-use loop. The AI has access to functions like queryExpertNetwork and proposeTerms. It autonomously decides which specialists to ping, handles their counter-offers, and presents the user with a "Winner."
• The Interface (React and Tailwind): Designed with a modern aesthetic—utilizing glassmorphism and a cinematic navigation dock for a high-end, infrastructure-management feel.

Challenges I ran into

• The Agentic Recursive Loop: Initially, the AI would call a tool but would not know how to wait for the result. I had to architect a multi-turn chat system that simulates a live API handshake, feeding contractor responses back into the model until a final contract JSON is produced.
• Visual Ambiguity vs. Technical Accuracy: A blurry photo of a boiler could be many things. I refined the system instructions to ensure Gemini 3 prioritizes safety. If the AI detects ambiguity in a high-voltage or gas-related image, it refuses the triage and requests a high-resolution video or immediate specialist intervention.
• Latency Balancing: High-reasoning tokens are incredible for negotiation but slower for simple UI searches. I implemented a tiered reasoning flow: Gemini 3 Flash handles lightning-fast UI interactions, while Gemini 3 Pro is only invoked for the technical diagnosis and agentic negotiation.

Getting Started

Prerequisites

• Node.js (v18.0.0 or higher)
• npm or yarn
• A Google Gemini API Key (obtained via Google AI Studio)

Installation

1. Clone the repository:

git clone https://github.com/yourusername/orion-home-ai.git
cd orion-home-ai

2. Install dependencies:

npm install

3. Set up environment variables: Create a .env file in the root directory and add your API key:

GEMINI_API_KEY=your_api_key_here

4. Start the development server:

npm run dev

What's next for Orion

• Live API Integration: I plan to integrate the Gemini 3 Multimodal Live API to allow users to walk through their house and talk to Orion in real-time, letting the AI hear the sound of a rattling HVAC unit.
• IoT Predictive Maintenance: Connecting to smart water meters and circuit breakers to let the Orion Negotiator proactively fix issues before the user even notices a leak.
• Global Grid Expansion: Moving beyond the local hub to integrate building codes from cities worldwide into the diagnostic engine.