- Over 800,000 job-seekers each year interview on platforms like HireVue and Sonru. Once a question is displayed, users typically have 30-60 seconds to prepare, then 1-2 minutes to record an answer.
- Once their responses have been submitted, an AI filters out candidates. Many never see human eyes.
Candidates receive no feedback.
We believe this is unfair and leads to many capable candidates being rejected for arbitrary reasons, such as not making eye contact with the camera.
HireYou seeks to level the playing field by allowing candidates to practice under timed conditions, with smart feedback powered by Microsoft Azure's Face API.
What it does
The core functionality of HireYou is to allow users to do many practice interviews in their own time. These interviews would mimic the flow of a HireVue session, except that users would be able to review their answers afterwards.
However, using the power of Microsoft Azure's Face API, we are able to enhance this concept by providing smart feedback. Concretely, we offer the following:
- The HireYou Confidence Score
- 8 dimensions of emotion, displayed dynamically
- An eye contact index
- Summary statistics
How we built it
We went with a simple NodeJS/Express/MongoDB stack with Nunjucks for templating and Semantic UI for styling, and WebRTC to record the video.
Challenges we ran into
- Azure rate limiting meant that we had to generate snapshots of the video on the fly instead of postprocessing them from the completed video
- When recording the video, our webapp is also displaying it, and taking snapshots of frames at a regular interval of one second to send to Azure Face API and save the result to our database. This was challenging for a number of reasons. We had problems recording the video and had to switch the library we used to do so, which made it impossible to use another library, Webcam.JS to take the snapshots. So we had to code the snapshot functionality in ourselves. Thankfully, setting up Microsoft’s Azure API was pretty simple, and overall, we fixed these problems.
- Later, we ran into a problem with our timers being affected by a global state and spent a few hours trying to fix it and recode some parts of our program to not rely on this state.
- Setting up MongoDB proved to also be a challenge with some initial stability issues
Accomplishments that we're proud of
- Overall question flow and user interface
- Completely functional with no hard-coding, and easily extensible.
- Emotion detection: provides credible results!
- It is a clear solution to a real problem: we will personally be using this app to prepare over summer
What we learned
- Time management: focus on core functionality before making minor tweaks
- User interface is very important!
What's next for HireYou
We seek to release HireYou into production by Michaelmas 2019, ready for the job application season. We will leverage our personal networks as well as the University Careers Service to drive adoption, and we strongly believe that word of mouth will be a powerful factor.
We would likely adopt a permissive freemium model in which users can have unlimited practice interviews, with a reasonable pricing scheme providing access to advanced features such as eye contact detection and emotion detection.
Specific features we would like to implement in the long run:
- Natural language processing on what the candidate is saying
- Allowing candidates to choose question banks based on specific companies
- Smarter algorithms for question selection