ClimaCrete Systems

Climacrete Systems is a ML-driven platform that helps builders and engineers design low-carbon concrete mixes without compromising performance.

Comment

Inspiration

Concrete is the world’s most used building material — second only to water. But the cement that binds it together is responsible for nearly 7–8% of global CO₂ emissions. Climacrete turns climate data into design intelligence — helping the construction industry build stronger, greener, and smarter. Inspired by climate research and industry need research in Design Climate I course

What it does

By combining real concrete durability data, material data, and aggregate data with machine learning, Climacrete predicts compressive strength, durability, and CO₂ impact — instantly showing the trade-offs between sustainability and structural performance. With an input of mix proportions or recycled materials, the app recommends optimal use cases and visualizes how supplementary materials like fly ash or slag affect both strength and carbon footprint.

How I built it

Gathered data from:

  1. Concrete Compressive Strength
  2. Natural fiber-recycled aggregate concrete compressive strength with web-based optimized UI

Used notebooks to clean & train models that were then outputted as .pkl files:

  • ├── notebooks/
  • │ ├── 01_data_cleaning.ipynb
  • │ ├── 02_training_strength_model.ipynb
  • │ └── 03_training_durability_model.ipynb

Used Streamlit to make a webapp & allow users to test and display the predictions

Challenges we ran into

  1. Cleaning the data & attempting to understand what it means (used LLM to help create formulas)
  2. Training the models
  3. UI

Accomplishments that we're proud of

Strength Model:

  • R²: 0.935
  • MAE: 3.18 MPa

Durability Model:

  • R²: 0.994
  • MAE: 0.39

What we learned

More practice with ML RandomForest

What's next for ClimaCrete Systems

Verifying that the formulas created and data cleaning are as scientific as possible and truly provide more than 75% accuracy Testing mixes in real life and comparing

Built With

Share this project:

Updates