HireEqual

Inspiration

Companies like HireVue and JobFair are utilizing AI facial and audio algorithms to automate the interview process and better understand and assess candidates. The issue with these systems is the lack of diverse candidate training data (which could cause bias in automated scores given to candidates who belong in underrepresented groups) and transparency in the algorithms used to evaluate the candidate during the automated interview process. Another common problem with these systems is the absence of personal touch--there is currently no intuitive way to answer the candidate's questions. To address these issues, I created HireEqual, a platform that functions similarly to HireVue but reduces algorithmic bias through StyleGAN and is more transparent and user-friendly.

What it does

HireEqual is a candidate assessment tool that anyone can take and see how they score. It assigns a score from 1-10 (1 being negative, 10 being positive) based on a model that detects positive and negative emotions (planning to expand to include a more diverse range of emotions). The model is open-sourced so there is transparency in how the algorithmic decisions are made and users can submit feedback if HireEqual's scores don't align with their actual emotion. HireEqual also uses StyleGAN to create an enriched, synthetic dataset based on candidate data and other datasets in order to train the model to ignore features such as skin color, gender, and other discriminatory factors.

How we built it

Python, tensorflow in the backend for model development

Challenges we ran into

converting it into a more usable product, but from an IDE standpoint, it is usable

Accomplishments that we're proud of

This was a huge project and with the limited time available, I was proud to complete as much as possible

What we learned

Backend + ML development

What's next for HireEqual

Address possible overfitting of the model, make HireEqual work for video data in addition to image data.

Built With

• python
• tensorflow