Oxford EDI Initiative

Overview

This project set out to tackle a powerful question: What if interviewers couldn’t see or hear any personal characteristics of a candidate — only their words?

The team designed a blind interview plug-in that replaces candidates’ faces and voices with a single, neutral model during video calls. By removing cues such as gender, ethnicity, and accent, the system aimed to reduce unconscious bias and create a fairer, more inclusive interview process.

Objectives 

• Develop a video call plug-in to neutralise personal characteristics in real time.

• Explore how technology can reduce unconscious bias in recruitment and assessment.

• Test the system in collaboration with the MPLS EDI Committee.

Approach

The team built the system using:

• Open Broadcaster Software (OBS) for video streaming.

• Face- and voice-changing plugins, integrated with custom-trained models.

• Generative Adversarial Networks (GANs) such as StyleGAN, fine-tuned through latent space editing.

• Gender and race classifiers to ensure balanced outputs.

After curating a diverse dataset, the models were trained and tested, with attention given to latency, processing power, and compatibility. A pilot study then put the prototype into practice, providing valuable real-world feedback.

Outcomes

• Delivered a working prototype of a blind interview plug-in.

• Demonstrated how AI and software can be applied to promote fairness in recruitment.

• Sparked meaningful discussion on the role of engineers in designing systems for social responsibility.