BiasShield

We were driven by the need to address bias in hiring — especially how tone, accent, or hesitations can unfairly affect candidates. Inspired by research from Moritz Hardt and Hoda Heidari, we built a tool that not only detects bias but helps correct it.

What it does

iasShield is an AI-powered fairness audit tool for interviews. Users upload a voice or video response. The system:

Transcribes using Whisper

Simulates 3 reviewers with GPT-4:

Neutral HR

Biased evaluator

Fairness auditor

How we built it

Streamlit frontend for fast UI

Whisper for speech-to-text

GPT-4 with prompt-engineered personas

Bias score generated by comparing persona reviews

(Optional) Emotion analysis using OpenCV + DeepFace

Entirely built in Python, deployed on Streamlit Cloud

Challenges we ran into

API intergration

Accomplishments that we're proud of

Adding researched frameworks

What we learned

What's next for BiasShield

Eastern Communtiy

Built With

• numpy
• python-streamlit-openai-whisper-api-openai-gpt-4-opencv-deepface-pandas