Vervoe logo

4 min read

Introducing Active Learning for Vervoe

Right out-of-the-box, our AI grades your candidates automatically with 80% accuracy. But we also recognize while building Vervoe that not all roles are created equal, and every organization has its personal hiring preferences.

Today we released Active Learning for Vervoe. This release is a special application of our AI that helps simplify training Vervoe to your preferences and make it even easier to find the best person for your role. 

Why we did this

Standard test grading isn’t always enough. Every company needs something a little different from candidates. For example, two companies might be looking for front-end developers (FED). Both require excellent technical and problem-solving skills. So both might use a similar test for the role.

But your company is looking for the only FED in the team, so the person you hire needs to communicate effectively with lots of different stakeholders. The other company might be hiring a FED that’s going to be part of a much larger team, so they’re looking for a coding superstar.

We knew we could make it easier for every organization to train our AI to their preferences. 

Our approach

In the past, you might have achieved a similar outcome by manually grading a bunch of candidate responses each week. Nobody wants to waste time on candidates that are a poor fit, so you probably would have manually graded just the high performers. However, to know what a “good” answer is, our AI must also understand what a “bad” answer is.

Vervoe’s AI grading engine (nicknamed autograder) uses the scores of one set of candidates to predict the scores for the next set of candidates; this is called the training set. 

Introducing active learning for vervoe 1

The training set should be balanced, meaning enough of both bad and good answers, so the autograder has a complete picture. Our new assessment grading feature will surface candidate responses for you to grade, to give the autograder this balanced dataset. 

Initially, Active Learning selects one random answer from each question in your assessment and then searches for the most opposite answer. This process continues until there’s a solid group of responses that are all as different as possible. These responses are then surfaced to you under Grading where you can give them a quick poor, average or excellent score. This becomes the new training set that “autograder” feeds off.

In short, if you value things like correct spelling or grammatical consistencies, you can train the AI to also value these by giving responses with spelling errors a lower grade. Or maybe you want to hire candidates that align with your core company values, so giving higher grades to answers that reflect more of these values will also train the AI to grade them more favorably.

What makes this feature unique?

You’ve probably encountered Active Learning through the Spam filter on your Gmail account. Each time you mark an email as ‘not spam’ or vice versa, you’re actively training the AI to understand what is and isn’t spam to you. The more newsletters, publications, communities, and other emails you subscribe to, the harder it becomes to manage what you see and prioritize the things that matter most to you. The personalization Active Learning enables simplifies the way you interact with the product and effectively manages scale.

Our Active Learning for candidate grading works in much the same way, using simple interactions to tell the AI which answers you prefer, and which ones you don’t.

Even though you’ve seen something like Active Learning before, you’ve never seen it used for recruiting. It’s a unique application that’s at the intersection of people and technology. You can truly interact with our AI and make it your own.

Active Learning has been six months in the making, and this release marks the beginning of some continuous improvements to our AI and the way you interact with it. As a team, we’re obsessed with finding the optimal amount of answers that need to be graded. We are haunted by this magic number and where we have to stop.

Want to see it for yourself? Get a free trial of Vervoe to start inviting candidates and see how Active Learning can work for you.

Anatolii Kryvko

Anatolii Kryvko

Data Scientist @ Vervoe

Similar articles you may be interested in​

Wish you could measure skills that directly lead to more revenue, find hidden gems and avoid unqualified candidates? Our new

May 17, 2023

We’re thrilled to announce the launch of our new embedded Excel based question type, updated integrations and more ways to

April 23, 2023

The year might have only just commenced, but we’ve been busy ensuring that Vervoe’s core technology remains innovative and ahead

March 1, 2023