CORROBORAI

Inspiration

Our goal was to create a tool that could understand the context behind an discrepancy, apply business rules, explain its reasoning, and help analysts focus only on the cases that truly deserve attention.

What it does

CORROBORAI compares employee data between System A, the HR source system, and System B, the time-management system.

How we built it

We built CORROBORAI in Python, using pandas for data processing, scikit-learn for the machine-learning components, and Streamlit for the interactive interface.

Challenges we ran into

One of the biggest challenges was realizing that a raw mismatch did not necessarily represent an error. Dates could be stored in different formats, fields could be transformed through business rules, and some destination values could not be compared directly to a single source field. We had to carefully separate normalization differences, legitimate rule-based transformations, and actual data errors.

Accomplishments that we're proud of

We managed to implement an AI in our project without prior experience

What we learned

systems. A technically different value may still be perfectly valid, while a seemingly small discrepancy can represent a real operational problem. We also learned that AI is most useful when it is given a specific role rather than being forced into every part of the system. Deterministic rules were better for cases with clear business logic, while machine learning was more useful for ambiguous matching, prioritization, and identifying patterns across the dataset.

What's next for CORROBORAI

The next step would be to test CORROBORAI on larger and more varied datasets to evaluate how well the matching and prioritization models generalize beyond the supplied challenge extract. We would also like to expand the reviewer-feedback system so confirmed human decisions can help improve future reconciliation rules and model training.