Aura Home AI: Autonomous Intelligence & Orchestration Hub

Autonomous home orchestrator that senses, reasons, and acts. Powered by Gemini 2.0 & MongoDB MCP to resolve financial leaks, security threats, and energy waste with total privacy.

Comment

🏡 Aura Home AI: Autonomous Intelligence & Orchestration Hub

🌪️ Inspiration

The modern home has become an unmanaged enterprise. Between "subscription creep" costing households over $900 annually and "dumb" HVAC systems wasting 25% of domestic energy, the average person is losing agency over their household economy. We are no longer masters of our environment; we are overwhelmed by its maintenance. Our inspiration was to build a system that moves beyond the "Smart Home" chatbot and into a true Home Orchestration Node—an autonomous assistant that doesn't just answer questions but autonomously resolves the crises of modern life.

⚡ What it does

Aura Home AI is a voice-activated home co-pilot that senses, reasons, and acts. It is architected around the Aura Seven—a hierarchy of seven specialist agents:

  • Finance Sentinel (FIN): Autonomously audits subscriptions and disputes overcharges.
  • Guardian Protocol (GRD): Reasons across vision feeds to secure the home perimeter.
  • Energy Architect (NRGY): Optimizes utility usage against live peak-rate data.
  • Pantry Pilot (PNTRY): Monitors inventory and reroutes orders for local value.
  • Wellness Warden (WLNS): Tracks family health metrics and automates prescriptions.
  • Time Weaver (TIME): Orchestrates complex schedules and automates the household "mental load."
  • Vision Advisor (VIS): Uses Gemini 2.0 Vision to provide live contextual home analysis.

All actions are visualized in a high-fidelity Glassmorphic Command Center and persisted in a private, secure MongoDB Confidential Vault.

🛠️ How we built it

  • AI Core: Built on Gemini 2.0 Flash via Vertex AI for high-speed, low-latency multimodal reasoning.
  • Orchestration: Leveraged the Google Agentic Design Kit (ADK) to manage the multi-agent hierarchy.
  • Confidential Data Grounding: Implemented the MongoDB Model Context Protocol (MCP) to ensure all decisions are grounded in the user's private, encrypted data vault.
  • Frontend: Next.js 16 (Turbopack) with Vanilla CSS and Framer Motion for a premium, high-fidelity UI.
  • High-Performance Telemetry Stream: Engineered an asynchronous state logger and active event emitter to feed real-time agent updates directly into our UI.

🏗️ Architecture

graph TD
    %% Styling
    classDef core fill:#0f172a,stroke:#38bdf8,stroke-width:3px,color:#fff,font-weight:bold
    classDef agent fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fcd34d,rx:10px,ry:10px
    classDef input fill:#334155,stroke:#10b981,stroke-width:2px,color:#a7f3d0,stroke-dasharray: 5 5
    classDef output fill:#334155,stroke:#ef4444,stroke-width:2px,color:#fecaca

    %% Core Components
    C[Aura Home AI<br/>Command Center / Orchestrator]:::core

    %% Inputs
    subgraph Sensors [Multimodal Sensory Inputs]
        S1(Vision Diagnostics / CCTV):::input
        S2(Utility API Feeds):::input
        S3(Financial Ledgers / Plaid):::input
        S4(IoT / HVAC Telemetry):::input
    end

    %% 7-Agent Autonomy Engine
    subgraph Autonomy [7-Agent Autonomy Engine]
        A1[Finance Sentinel <br/>FIN]:::agent
        A2[Guardian Protocol <br/>GRD]:::agent
        A3[Pantry Architect <br/>PNTRY]:::agent
        A4[Energy Optimizer <br/>NRGY]:::agent
        A5[Wellness Advisor <br/>WLNS]:::agent
        A6[Vision Advisor <br/>VIS]:::agent
        A7[Timeline Coordinator <br/>TIME]:::agent
    end

    %% Outputs
    subgraph Actions [Autonomous Resolutions]
        O1(Dispute / Refund Submissions):::output
        O2(Grocery Rerouting / Ordering):::output
        O3(Thermostat Adjustments):::output
        O4(Perimeter Threat Alerts):::output
    end

    %% Connections
    S1 --> |Video Stream| A6
    S2 --> |Rates| A4
    S3 --> |Transactions| A1
    S4 --> |Temp/Air Quality| A5
    S1 -.-> |Motion| A2

    A1 <--> C
    A2 <--> C
    A3 <--> C
    A4 <--> C
    A5 <--> C
    A6 <--> C
    A7 <--> C

    %% Cross agent logic
    C --> |Coordinate 38+ Decisions| A7

    A1 --> |Execute Audit| O1
    A3 --> |Market Shift| O2
    A4 --> |Load Balance| O3
    A2 & A6 --> |Security Breach| O4

🚧 Challenges we ran into

Building a system that "sees and hears" simultaneously across multiple domains presented significant hurdles:

  • Architectural Sync: Coordinating seven distinct specialist agents while maintaining sub-second response times required strict prompt engineering and state management.
  • Multi-Agent Latency Calibration: Preventing thread contention and optimizing Vertex AI streaming response times when seven specialist agents are triggered concurrently by a single complex query.
  • Grounding Data Integrity: Ensuring the model never "hallucinates" clinical or financial data required a robust, secure tool-callback layer via the MongoDB MCP.

🏅 Accomplishments that we're proud of

  • The Aura Seven: Achieving perfect intent routing across seven highly specialized household domains.
  • Zero-Click UI: Developing a "Command Center" that feels alive with background autonomous pulses even when the user isn't interacting.
  • Dynamic Event Streaming: Building a live telemetry visualizer that cleanly exposes every underlying agent decision and API callback directly to the operator's viewport.
  • MongoDB Integration: Successfully grounding the agentic reasoning in a persistent, secure private vault using the Partner Track MCP.

📖 What we learned

We learned that the true power of AI in the home isn't "Conversation"—it's Resolution. A model that can tell you that you're wasting money is a chatbot; a model that can stop the leak is an agent. We also learned the critical importance of Data Autonomy; for an agent to be trusted with home life, it must operate on a private, persistent vault (like MongoDB) that the user truly owns and controls.

🔭 What's next for Aura Home AI

  • FHIR Healthcare Integration: Connecting the Wellness Warden to real, compliant patient records for automated caregiving.
  • Hardware Node: Moving Aura from the browser into a dedicated, privacy-focused hardware device—the Aura Home Core.
  • Expansion of the Vault: Extending the MongoDB MCP layer to support decentralized, local-first data persistence for maximum privacy.

Built With

Share this project:

Updates