Presage x AI Interview Pressure Test

Our initial Menu Screen
The chat window before the data loads & questions are asked
The chat window once the interview begins and biometric data is read in
Our final review, analysis and coaching screen

Inspiration

One of our team-members was struggling with job interviews- bombing one the night before Owl Hacks. We realized that college students needed a better way to practice conducting job interviews-without the help of other people.

What it does

Our application gives the user the opportunity to conduct simulated interviews. Once the program starts, the user is greeted with a menu- giving them the opportunity to input the job title and description they wish to practice interviewing. This exists alongside the option to select various interviewer personas (Friendly, Harsh and Quick Paced).

Once the user enters the simulated interview, the program gathers their resting heart rate through Presage (in both chart form and categorizing the user as either calm or stressed) and the interviewing Agent (a Gemini persona with ElevenLabs voice integration) asks the first candidate specific question. Once the user responds- either via microphone or text- the interview session runs until completion where it offers comprehensive data (including which questions spiked their heart rate the most), analysis and coaching.

How we built it

We utilized Claude Code and Cowork to build a three-stage Node.js application. We have a back-end (processing data, determining whether the user is stressed and using our AI API calls), a front end (showing the user charts, a GUI and their chat window) and a 'bridge' to handle our API calls to Presage. We put a lot of work into managing API keys, working on the Gemini prompts for the interview personas and fine tuning our GUI for the most intuitive, aesthetic and user friendly experience possible.

Challenges we ran into

Our largest challenge came with integrating our biometric data from Presage. It took us several hours of troubleshooting to figure out that the Codespace environment, in our case, was incompatible with the web-camera streaming required for our Presage usage. To address this, we installed Node.js locally and ran the backend 'Presage-Bridge' and 'Front-End' on a Windows laptop.

Accomplishments that we're proud of

We are proud to have built a functional application complete with biometric data (something we feel is rather novel for this type of software)- addressing a real problem we have encountered. We are happy with both how the voice integration ran seamlessly and the quality of responses and analysis provided by Gemini- creating a very interactive experience for the end user.

What we learned

As our first Hackathon, we learned a lot about fundamentals such as how to branch & merge on Github, use Cowork and Node.js commands. We used Github's Codespace for the first time and found it to be an enjoyable and seamless experience (despite its potential limitations as per our Presage integration issues).

Outside of this, we learned a lot about APIs, working with AI Integrations and how crucial version control is on Github.

What's next for Presage x AI Interview Pressure Test

We want to refine Gemini's responses further, improve which ElevenLab's voices we use based on the interviewer persona and overall just make the platform more user friendly for our own use. If the project becomes popular, we'd love the chance to release it to the wider public.