From Hackathon to Impact: Duke Students Build SAFE-T to Audit Bias in Transportation AI

SCAI Hackathon winners advance progress with SAFE-T development

What began as a weekend prototype at the Society-Centered AI Hackathon is now gaining traction as a tool to evaluate fairness in how cities plan their transportation systems.

SAFE-T (Safety Algorithm Fairness Evaluation for Transportation) was developed by a team of Duke AI Master of Engineering students during the SCAI Hackathon, where it earned second place. The project set out to answer a critical question: are the AI systems used in transportation planning treating all communities equitably?

Cities across the United States increasingly rely on AI-driven tools such as traffic volume counters, crash prediction models, and infrastructure planning systems to decide where to build sidewalks, crosswalks, and bike lanes. These tools play a major role in directing billions of dollars in safety funding. However, evidence suggests they may systematically undercount pedestrians and cyclists in low-income and minority neighborhoods.

The SAFE-T team used Durham, North Carolina as a pilot case to test their framework. Their audit revealed a 38-percentage-point gap in how accurately AI systems identify high-risk neighborhoods in low-income areas compared to wealthier ones. In a city where 47 percent of pedestrian and cyclist crash victims are Black residents, but Black residents make up 32 percent of the population, the findings point to meaningful disparities in how risk is measured and addressed.

Since the hackathon, the project has continued to evolve. The team presented SAFE-T at the SCAI 2026 Conference and is now in discussions with the City of Durham Transportation Department about integrating the framework into real planning workflows. They are currently applying for funding to scale SAFE-T into a tool that could be used by cities nationwide.

SAFE-T remains fully open source, with a live dashboard available to the public: safe-t-ai.github.io.

The project reflects the core goal of the SCAI Hackathon: to move beyond ideas and build systems that engage directly with societal challenges.

The team includes Lindsay Gross, Jonas Neves, Arnav Mahale, and Shreya Mendi.