SPARKnSMART (SNS) - Appliance-Level Energy Tutor

Project Overview

A messaging-first, multi-agent tutor that teaches appliance-level energy concepts and actionable savings. Built with Google Antigravity and the Agent Development Kit (ADK).

Tracks

• Primary: Agents for Good
• Secondary: Concierge

Architecture

The system uses a multi-agent architecture with the following agents:

• AdvisorAgent: Generates educational content using Gemini.
• AnalyzerAgent: Analyzes appliance data for cost and power factor issues.
• ReporterAgent: Creates visual charts of energy usage.
• NotifierAgent: Formats and sends messages via Matrix.
• SessionAgent: Manages user session state.

Setup

1. Clone the repository:

   git clone <repo_url>
   cd energy_tutor

2. Install dependencies:

   pip install -r requirements.txt

3. Configure Environment: Create a .env file with the following:

   GEMINI_API_KEY=your_gemini_key
   MATRIX_USER=@your_user:matrix.org
   MATRIX_PASSWORD=your_password
   MATRIX_ROOM_ID=!your_room_id:matrix.org

4. Run the Demo:

   python demo_push.py

Evaluation

The system includes offline scenarios to verify functionality.

• Scenarios: src/data/samples/ contains appliances.json and tariff_telangana.json.
• Metrics: Latency, bill estimates, and action counts are logged.

Deployment

Build and run with Docker:

docker build -t energy-tutor .
docker run --env-file .env energy-tutor