Opportunity XplAIn

Opportunity XplAIn is an AI-powered tool that predicts whether a customer is likely to be profitable based on their past interactions with the company.

Comment

Inspiration

For the Schneider Electric Datathon 2025 challenge, we wanted to create a solution capable of helping businesses anticipate whether a customer is likely to convert in a sales opportunity. Our goal was to combine machine learning with explainability to make predictions not only accurate, but also transparent and easy to understand for non-technical users.

What it does

Opportunity XplAIn is an AI-powered web application that predicts whether a customer is likely to be profitable in a planned sale based on previous interactions. The system includes:

  • A binary classification model achieving ~0.75 accuracy and F1-score.
  • A second LLM model that interprets SHAP values and generates a human-readable explanation of why the prediction was made.
  • A user-friendly interface displaying both AI-generated explanations and intuitive graphs to help users understand the model’s reasoning.
  • Two data-entry modes for flexibility: percentage-based inputs and interactive sliders to adjust each variable visually.

How we built it

We built the system using a modern full-stack architecture:

  • Machine Learning: Python, TensorFlow, SHAP, and interpretability techniques.
  • Backend: FastAPI to expose prediction and explanation endpoints.
  • Frontend: React + TypeScript with a custom UI for interacting with the model and visualizing results.
  • DevOps: Git, GitHub, Docker, and Linux for development, version control, and containerization.

The ML model was trained on customer opportunity data, and then integrated with a secondary LLM (via Ollama) to generate structured natural-language explanations.

Challenges we ran into

  • Ensuring TensorFlow compatibility and stable environment configuration.
  • Managing SHAP outputs and translating them into readable explanations.
  • Designing a smooth user experience that balances visual clarity with technical detail.
  • Integrating two different AI components (ML model + LLM interpreter) within a unified backend.
  • Handling input variations while keeping the interface intuitive and responsive.

Accomplishments that we're proud of

  • Achieving strong predictive performance (~0.75 F1-score and accuracy).
  • Successfully integrating SHAP with an LLM to produce high-quality explanations.
  • Building a clean, modern web UI that makes complex ML results understandable.
  • Delivering a fully functional end-to-end system during the Datathon timeframe.
  • Designing flexible input mechanisms that simplify the user workflow.

What we learned

  • Effective ways to combine traditional machine learning with large language models.
  • Best practices for ML interpretability and SHAP visualization.
  • How to build a scalable backend architecture using FastAPI.
  • How to design a polished, intuitive UI for data-heavy applications.
  • The importance of explainability and transparency in AI decision-making.

What's next for Opportunity XplAIn

  • Adding more granular interpretability methods (e.g., feature interactions, waterfall plots).
  • Improving prediction calibration and threshold optimization.
  • Expanding the UI with richer visualizations and report-generation capabilities.
  • Deploying the system as a fully cloud-hosted service for real-world testing.
  • Exploring integrations into Schneider Electric’s internal workflow tools.

Built With

Share this project:

Updates