Stelf AI

Inspiration

Buildings contribute significantly to global energy consumption. The inspiration behind Stelf AI is to address this challenge with a practical solution that empowers operators to make data-driven energy decisions while supporting decarbonization goals.

What it Does

Stelf AI provides:

• Short-term load forecasting with pre-trained TSFMs
  
• Anomaly detection for unusual consumption patterns
  
• Fine-tuning for specific building use cases
  
• A web-based dashboard with interactive time-series visualizations
  
• Insights and recommendations for energy optimization
  

How We’ll Build It

We will integrate pre-trained TSFMs with a backend API for predictions, connected to a React-based dashboard for visualization. Data will be stored in a scalable database, and optional fine-tuning will be enabled using PyTorch for specific building contexts.

Challenges

• Handling variability across different building types
  
• Designing a dashboard that is both powerful and simple for non-technical users
  
• Ensuring forecasts remain accurate under noisy or incomplete data
  

Accomplishments We Aim For

• A forecasting dashboard that is intuitive yet technically robust
  
• Seamless integration of pre-trained models with optional fine-tuning
  
• Real-time anomaly alerts for building operators
  

What We Learned (Expected)

We expect to gain experience in applying foundation models to real-world time series, combining machine learning with full-stack development, and addressing user experience in technical dashboards.

What’s Next for Stelf AI

We plan to extend Stelf AI into a scalable SaaS product that supports multiple building types, integrates IoT sensor data, and enables energy providers to manage demand response more effectively.

Built With

• express.js
• mongodb
• node.js)
• pandas
• pytorch
• react.js
• scikit-learn