Re-Estate

Your fully automated AI Real Estate agent. Predict the best price range, generate memorable titles and descriptions to market a property, and analyze its surrounding regions.

Comment

✨ Inspiration

Never undersell the memories you made in your home.

Using our platform, everyone can insert the features of their property to get an accurate estimate of the real estate features in comparison to others in the region, then automatically generate a catchy title and description to market it to the public.

❓ What it does

  • Our project uses the Restb.ai dataset to predict the price of a property based on its features (size in square meters, number of rooms, ...) and then how this price can be modified by adding other features to it.
  • Users can insert their own features to get a price range based on the AI model
  • A heatmap of the average prices per neighborhood can be plotted in the "Value Heatmap" tab
  • You can also explore the relevant features and explainable AI methods with an interactive sidebar
  • Generate title and text for the real estate property using OpenAI's API
  • The average sale per month (€) in each region in Spain is shown in the last tab

Go to https://mlheads.streamlit.app to check out our working prototype or just click the following button: Streamlit App

💻 How we built it

  • Technologies: Domain.com, SHAP, Streamlit, Counterfactual, Git
  • Datasets: Restb.ai's API, OpenAI's API, OpenStreetMap
  • Language: Python, HTML, CSS

⚔️ Challenges we ran into

  • Loading the complete dataset took some time to solve
  • Coming up with a creative, innovative, and useful idea
  • We spent a lot of time trying to integrate Auth0 and always felt like we were one step away from accomplishing the task, but then decided to leave it to future steps

🏅 Accomplishments that we're proud of

  • We managed to clean and preprocess 10.000 rows of Restb.ai's dataset for training and validation
  • Developed two explainable AI algorithms with SHAP and Counterfactual to assess the relevant features
  • We learned from 0 how to make an explainability of the models and turn uncertainties into explanations
  • Created our own original dataset of coordinates (latitude and longitude) of provinces in Spain for visualization
  • Built a working frontend-backend pipeline with an interactive user interface that has been deployed online
  • Collaborated and had lots of fun together despite all the above challenges whilst completing our project in less than 36 hours!

📚 What we learned

The knowledge we have learned stands out not only for its quantity but also for its depth

  • We learned how to explain and intepretate a black-box model
    • SHAP (SHapley Additive exPlanations) is a method that assigns importance values to features in a model, providing insights into how each feature contributes to the prediction, aiding in interpretability and understanding of the model's decision-making process.
    • Counterfactual techniques involve generating hypothetical scenarios to understand the causal effects of different variables on an outcome, helping to explain why a particular outcome occurred or to explore "what-if" scenarios.
  • We had a good opportunity to broaden our understanding of AI and ML, especially since not all of us have a background in those fields.
  • We also got deeper into some aspects that we had some experience before like Git and Streamlit, utilizing the tools to their full extent with GitHub Project and various Streamlit integrations
  • It has been very interesting studying how a house's characteristics can affect its price and which of them are the most important.
  • We also extended our knowledge of APIs and Auth0 in our deep attempt to implement user authentication in our platform

🔮 What's next for Re-Estate

Re-estate has a lot of future ahead and a box full of ideas to implement and create a complete and innovative platform. Our goal is to achieve a virtual real estate assistant for all those sellers who need to sell, promote, or simply study their properties. Many times, they are not aware of the value they hold in their hands, and that's why Re-estate intends to maximize the capabilities of artificial intelligence to facilitate these processes that can be burdensome.

Our next steps involve creating a section where the user can study all the characteristics involved in the prices of houses in a specific area by themselves. Considering that during this challenge, we haven't had enough time or computational capacity to train cutting-edge models, another aspect to address would be to improve our models, as well as apply other explainability techniques to understand how our model acts in different ways. Finally, another pending task is to use images as a tool and input for our artificial intelligence models.

🤗 Meet our team!

  • Diaaeldin Shalaby (Egypt)
  • Lluis Llull Riera (Spain)
  • Gabriel Orbe (United States)
  • Nathanya Queby (Indonesia)

🗣️ Fun fact

Our GitHub repository's README.md file is available in three languages: English, Spanish, and French!

Built With

Share this project:

Updates