InterviewSensei

Landing Page
Interview dashboard
Post-interview feedback generated

Inspiration

Finding internships is one of the hardest parts about being a student. Going into interviews not knowing what to expect, having to stumble through interviews because there aren't enough resources for getting through your first interviews, and not knowing what technical problems you might be thrown for your role.

Of all the ways AI could be used, we thought it should be used in a way that makes the path easier for any student looking for their first jobs, anyone looking to enter a new industry, and anyone striving to gain access to new opportunities that could change their lives.

What it does

InterviewSensei is a professional interviewer that can carry out evaluations for a variety of roles. With only a click of a button, InterviewSensei will prompt you that the interview has started and ask a variety of classic interview questions like your experience, how you react to difficult work situations, and more.

After the interview has ended, InterviewSensei will generate a report evaluating your performance on the interview, offering tips and advice on how to improve for your next interview (whether that might be for your internship offer next week, or your next session with InterviewSensei in a few seconds.)

How we built it

It's an integration of WebSpeech API to transcribe the user's speaking voice, OpenAI's ChatGPT to respond and formulate interview prompts, and ElevenLabs' Text to Speech API in order for the interviewer to speak back out loud.

Challenges we ran into

We were concerned about latency in the response times the availability of adequately natural-sounding text-to-speech APIs, and the fine-tuning of the actual interview content as our primary issues. WebSpeech API was highly efficient to use and reduced latency, and after several different API trials, ElevenLabs proved to be the most natural sounding API without unreasonable lag time.

However, though the moving parts worked individually, integrating them end-to-end created a lot of issues that we had to iron out in our product.

Accomplishments that we're proud of

ChatGPT, though its text generation capabilities are incredible, wasn't designed to be an interviewer. By fine-tuning elements like a roadmap for required sections of an interview, the expected tone of interviews, recommended content to include, with countless more prompt engineering trials, we created an interviewer that we're proud of and that we hope to make even more advanced in the future.

What we learned

We landed on our idea relatively late into the hackathon, and form there it was an organizational rush of brainstorming, streamlining, delegating, and cutting out unnecessary features. Our current product is the most optimized version that we could have developed, and the process of creating that was an incredible experience. The prevalent use of APIs was a new dimension as well that we hadn't considered before, and we learned a lot about interacting with various APIs for the use of our product.

What's next for InterviewSensei

With more fine-tuning, we'll soon add the option of adding a resume for input, having distinct industry and role-specific interview styles, and choices between highly behavioral versus technical interviews.

There are countless aspects of interviews to optimize—we plan to add salary negotiation, to integrate VR to offer the experience of speaking to an actual person, and to begin evaluating confidence and other self-presentation factors. The possibilities and use cases for InterviewSensei are endless!