As we grow, our applicant pool does as well. Reviewing applicants can be costly if it is taking time from our Engineering Leads. With some simple logistical regression from all of the data in Greenhouse, we can weigh certain words based on how often that word was associated with a rejected or hired candidate.
What it does
We pool together all of a candidate's submission: CV, GitHub (number of stars, repositories, and followers), and LinkedIn profile. Based on all of our previous offers and rejections, we can weigh words in a point system, where words that occur often in strong candidates count for more, and words that occur often in a weak candidate, are negative.
How we built it
"By a simple logistic regression and some NLP" - Alex Tselikov
Challenges we ran into
How to get appropriate data
How to parse CV's that often come in PDF format
How to extract relevant keywords
How to present results
Accomplishments that we're proud of
It works! This point system suggests that we are most successful in hiring quality candidates when they are referred, or met at an event.
What's next for AI Hiring
Using a workable version in daily hiring practices as an augmented hiring tool. A tool like this is unlikely to fully replace human review because of biases the system can learn.