Inspiration Tax fraud reduces public funding for essential services such as healthcare, education, transportation, and emergency response. We wanted to explore how AI could identify suspicious tax-filing patterns while protecting honest taxpayers from unfair accusations, bias, and privacy violations.
What It Does TaxFraud is an explainable AI dashboard that identifies tax filings requiring additional human review. Using synthetic filing data, it detects unusual patterns such as inconsistent income, duplicate information, and unusually high deductions.
The system provides:
- Transparent risk scores
- Plain-language explanations for flagged filings
- A case-review workflow for tax professionals
- Fairness and false-positive metrics
- An appeal process for taxpayers
- Human tax-preparer review before any action is taken TaxFraud does not determine whether someone is guilty. It only identifies patterns that may require further investigation.
How We Built It We mainly used IBM Bob in addition to GitHub to help design and develop TaxFraud as an interactive web dashboard. The dashboard displays summary information, highlights potentially suspicious filings, and explains the factors contributing to each risk score.
Challenges We Ran Into One challenge was balancing fraud detection with fairness. A system that flags too many cases could unfairly burden honest taxpayers, while a system that flags too few cases could miss serious fraud. We also had to determine which information could be used responsibly. We avoided sensitive personal characteristics, used synthetic data, and worked to explain risk scores clearly to nontechnical users.
Accomplishments We’re Proud Of We are proud to have created a system that focuses not only on detecting suspicious activity but also on protecting taxpayer rights.
Our project:
- Makes AI recommendations understandable
- Supports human review instead of automatic punishment
- Measures potential bias and false positives
- Protects sensitive information
- Allows taxpayers to appeal
- Connects fraud prevention with protecting public resources
What We Learned We learned that responsible AI requires more than building an accurate model. A useful system must also be transparent, fair, secure, accessible, and accountable. We also learned that AI should support human decision-making rather than replace it, especially when decisions can affect someone’s finances, reputation, or access to public services.
What’s Next for TaxFraud Next, we would test the system with tax professionals using carefully protected and anonymized data. We would improve fairness testing and strengthen privacy and security controls.
Log in or sign up for Devpost to join the conversation.