GridLens

Inspiration-

Buildings account for one-third of global energy use and emissions. We were inspired by the need for smarter, greener energy management that reduces waste and integrates renewables seamlessly.

What it does-

GridLens forecasts short-term building energy loads using Time Series Foundation Models (TSFMs) and detects anomalies in real time.

Predicts next-hour consumption

Flags unusual spikes/drops

Helps optimize renewable usage and battery storage

How we built it-

Leveraged pre-trained TSFMs for generalizable time series forecasting

Fine-tuned the models for building-level load data

Built a dashboard to visualize forecasts, anomalies, and carbon impact

Integrated with Python (for ML pipeline) and React (for frontend)

Challenges we ran into-

Handling incomplete or noisy energy datasets

Optimizing TSFM fine-tuning for different building types

Ensuring real-time inference without high computational cost

Accomplishments that we're proud of-

Achieved accurate load forecasting without retraining from scratch

Built a scalable anomaly detection system adaptable to multiple building types

Created a clean, intuitive visualization for building managers

What we learned-

How foundation models can be applied beyond NLP and vision

Trade-offs between accuracy, speed, and scalability in forecasting

The importance of combining technical AI solutions with practical UI/UX for adoption

What's next for GridLens-

Expand to multi-building and city-wide forecasting

Integrate with IoT sensors for richer real-time data

Add a carbon footprint tracker that nudges users toward greener consumption

Offer GridLens as a plug-and-play API for smart buildings

Built With

• docker
• flask
• numpy
• pandas
• python
• tfsm