This repository contains the official code and presentation materials for the hands-on workshop presented at GDG Presidente Prudente - DevFest 2025 in .
AI models are often "black boxes." They give us predictions but can't explain why. This is dangerous—it can lead to biased, unfair, and untrustworthy systems.
In this workshop, we perform an "autopsy" on a black-box model. We will:
- Train a high-performance classifier on the classic Titanic dataset.
- Debug its logic to understand why it predicts survival or death.
- Use powerful Explainable AI (XAI) techniques LIME and SHAP to make the model transparent.
- Generate powerful visualizations to explain individual predictions and global model behavior.
The easiest way to run the workshop code is by using Google Colab, which runs everything in your browser for free.
Just click the badge below!
- Python 3
- Libraries: scikit-learn, LightGBM, Pandas
- XAI Tools:
lime,shap - Dataset: The Titanic
ai_autopsy_xai_titanic.ipynb: The main Google Colab notebook with all the hands-on code.slides/: A directory containing the workshop presentation slides (e.g., in PDF format).
- Mateus Mendonça Monteiro: [MSc student and Backend Software Engineer at Luizalabs]
- LinkedIn:
linkedin.com/in/mateus-men - GitHub:
github.com/mateusememe
- LinkedIn:
- Maria Beatriz França: [Data Scientist Lead in Vericode]
- LinkedIn:
linkedin.com/in/maria-beatriz-frança/ - GitHub:
github.com/mbsf
- LinkedIn: