Inspiration
"Tell me about your biggest weakness - how did you go about addressing it?" Such complex questions - a common part of most job interviews - tend to be difficult to answer on-the-spot, often leaving interviewees perplexed and anxious. It's almost impossible to accurately simulate an interview environment in preparation. Very few have access to objective expert interviewers in their field of work, and the emotions one expects to encounter during an interview tend to be vastly different from the emotions they actually experience under such pressure. PrepAlchemy sets out to build a solution - an AI expert for interview prep, equipped with emotional analysis capabilities to prepare for specific high-stress interview scenarios.
What it does
PrepAlchemy analyzes, in real-time, granular facial expressions alongside verbal audio input to provide users with support for anxiety reduction. It maintains a detailed log of chat history for user reference. At the same time, it provides interview feedback to users to support the revision process.
How we built it
We built our AI interviewer on visual studio code through the Hume AI api, configuring to roleplay 2 different potential interviewers - a strict and relatively relaxed interviewer. The interviewer provides real-time feedback focused on interviewee professionalism and answering ability. We also added hume ai's facial expression analysis to support users under extreme stress. After concluding the interview, users are provided with an analysis of their verbal emotions for reference.
Challenges we ran into
At the beginning, we had difficulty downloading the hume libraries into replit so we had to test out different text editing platforms to see what other options could remedy this situation. Ultimately, we settled on using vscode for the hume integration and replit for the front-end and audio capturing aspect. Other challenges that we ran into and are currently working on is the smooth integration of different platforms into a cohesive system (replit, vscode, hume ai, etc).
Accomplishments that we're proud of
We're really proud of how the different pieces of insights are comprehensively tied together with more work putting towards it. We also really like how when we tweak the instructions given to the human AI, it's becoming more empathetic to the user input while also allowing our users to simulate an actual interview environment that's both professional (with domain knowledge within the interview role) and accessible (more available to the interviewees to practice as practicing with a live human may require scheduling and a lot of other unpredictable hassels).
What we learned
We learned to integrate our knowledge within platforms such as visual studio code and replit, as well as using API keys to create custom chatbots.
What's next for PrepAlchemy
In order to simulate the interview experience in depth, we hope to incorporate an AR/VR component so the user can be placed in a more realistic interview scenario. It can also be extended to not only focus on interviewees, but also analyze interviewer behavior and allow them to be mindful of professionalism and ethics. The selection of interviewers can be extended to varying levels of strictness, alongside more diversity in interviewers demographically to remove biases. We also plan to embed a resume and portfolio upload option for interviewers to use information about the candidate to formulate detailed, specific questions.
Log in or sign up for Devpost to join the conversation.