⚡ Smart AI Energy Consumption Predictor

AI-powered system for predicting electricity consumption and providing personalized optimization recommendations using RAG, Agentic AI, and MLOps.

🎯 SDG 7 - Affordable and Clean Energy

---

🌟 Features

Core ML

• 📊 Energy Forecasting - ML models (Linear Regression, Random Forest, XGBoost)
• 🔍 SHAP Explainability - Transparent, interpretable predictions
• 🚨 Anomaly Detection - Identify unusual consumption patterns

AI/LLM

• 🤖 RAG-Powered Chat - Knowledge base with semantic search
• 🧠 Agentic AI - Google Gemini-powered energy advisor
• 💡 Smart Recommendations - Personalized optimization tips

Business Logic

• 💰 Bill Estimation - Indian tariff slab calculations
• 🌍 Carbon Tracking - CO₂ emissions with India factors (0.82 kg/kWh)

MLOps & Production

• 📈 MLflow Tracking - Experiment logging and model registry
• ⚙️ Optuna Tuning - Automated hyperparameter optimization
• 🐳 Docker Ready - Containerized deployment
• 🚀 FastAPI Backend - REST API for integrations

---

📁 Project Structure

sustainability/
├── api/                     # FastAPI backend
│   ├── main.py              # API entry point
│   ├── schemas.py           # Pydantic models
│   └── routes/              # Endpoint handlers
├── src/
│   ├── data/                # Data pipeline
│   ├── training/            # ML model trainers
│   ├── mlops/               # MLflow + Optuna
│   ├── anomaly/             # Anomaly detection
│   ├── agent/               # LLM/RAG components
│   ├── carbon/              # Carbon calculations
│   └── train_models.py      # Training orchestrator
├── dashboard/
│   └── app.py               # Streamlit UI
├── Dockerfile               # Container build
├── docker-compose.yml       # Multi-service setup
├── docs/                    # Documentation
│   ├── DECISION_LOG.md      # Technical decisions
│   ├── MODULE_MAP.md        # Module responsibilities
│   └── KNOWN_UNKNOWNS.md    # Limitations
└── ARCHITECTURE.md          # System design

---

🚀 Quick Start

Option 1: Local Development

# Install dependencies
pip install -r requirements.txt

# Generate data & train models
python src/data/data_generator.py
python src/train_models.py

# Launch dashboard
streamlit run dashboard/app.py

Option 2: Docker

# Run dashboard only
docker-compose up -d dashboard

# Run with MLflow
docker-compose --profile mlops up -d

# Run with API
docker-compose --profile api up -d

Option 3: API Server

uvicorn api.main:app --reload
# Swagger docs: http://localhost:8000/docs

---

📊 API Endpoints

Endpoint	Method	Description
/api/v1/predict	POST	Get consumption forecast
/api/v1/recommend	GET	Get optimization tips
/api/v1/carbon	POST	Calculate CO₂ emissions
/api/v1/carbon/quick	GET	Quick carbon lookup
/health	GET	Health check

---

🤖 ML Models

Model	MAPE	Status
Linear Regression	0.00%	✅ Baseline
Random Forest	2.71%	✅ Production
XGBoost	3.07%	✅ Production
LSTM	-	🔄 Planned

Target: MAPE < 10% ✅ Achieved

---

📈 MLOps Features

Experiment Tracking

# View experiments
mlflow ui --port 5000

Hyperparameter Tuning

# Optimize XGBoost
python src/mlops/optuna_tuning.py --model xgboost --n-trials 50

Anomaly Detection

python src/anomaly/detector.py
# Output: 438/8760 anomalies (5.0%)

---

🌱 Carbon Calculation

CO₂ (kg) = kWh × 0.82

India grid emission factor: 0.82 kg CO₂/kWh (CEA 2023)

---

🎯 Responsible AI

• ✅ Fairness - Unbiased datasets and recommendations
• ✅ Transparency - SHAP explanations for all predictions
• ✅ Privacy - No personal data storage
• ✅ Auditability - Rule-based recommendations, not black-box

---

📈 Expected Impact

Dimension	Impact
🌍 Environmental	15-25% energy reduction potential
👥 Social	Increased sustainability awareness
💰 Economic	Lower electricity bills

---

📄 License

MIT License - Built for SDG 7: Affordable and Clean Energy

---

Developer: Kaustubh Agrawal | Manipal Institute of Technology, Bengaluru

Built with ❤️ for a sustainable future