AI Interview with Celia

Login page
Paste job description
Voice-based behavioral interview session
Get feedback!

Inspiration

Facing the hardships of the job market in 2024, four software engineering students at SJSU decided to create Celia. They realized that there were many technical job interview platforms; however, there is a lack of behavioral job interview platform. After this realization, at SiliconXHacks 2024, the four SE students decided to create Celia, an AI that simulates a behavioral job interview and interacts with the user in real-time.

What it does

Celia, an innovative tool that uses GPT-4 to simulate the job interview process with impressive realism. With Celia, you can get custom interview questions based on the specific job description you provide, making the experience highly relevant and tailored to your needs. The AI interacts with you using voice, asking questions just like a real interviewer would, and even following up based on your previous answers and the job details. Plus, you can easily review your past interactions thanks to the detailed message history. If you want feedback on your performance, Celia can generate a comprehensive report from the entire interview, offering valuable insights to help you improve.

How we built it

Our web application combines modern web technologies and tools for a seamless user experience. For the front end, we used React+Vite, MUI, and Tailwind CSS to create an appealing interface. Firebase handles authentication, allowing users to log in and start their interview practice sessions easily. Users can copy and paste job listings from sites like LinkedIn and ZipRecruiter to begin their preparation.

To provide a realistic interview experience, we leverage the OpenAI API with advanced models like GPT-4 Turbo. When a user provides a job description, this information is used to generate relevant interview questions. During the interview session, user responses are processed by our fine-tuned AI, ensuring contextually appropriate and dynamic follow-ups.

We simulate real-time interviews using FastAPI combined with Text-to-Speech (TTS) and Whisper models. The TTS model converts text questions into spoken words, while the Whisper model transcribes spoken responses back into text. JavaScript audio blobs manage these interactions, ensuring smooth communication. All user responses and interview histories are securely stored in Firebase. After each session, users can generate detailed feedback on their performance. On the backend, we use Python with FastAPI for data processing, API interactions, and storage operations.

Challenges we ran into

To simulate a real-time interview session, we initially planned to use WebSocket for two-way communication between the frontend and backend. However, due to time constraints and the complexity of implementing WebSockets within our 36-hour timeframe, we had to find an alternative. We discovered that using JavaScript audio blobs along with GPT's Text-to-Speech (TTS) and Whisper models allowed us to achieve our goal effectively. By leveraging these tools, we realized that we could use a RESTful API instead, simplifying the implementation while still providing a seamless real-time interview experience.

Accomplishments that we're proud of

We are proud that we were able to get the core functionality of Celia working before the hackathon ended. In particular, we are most proud of making Celia interactable and fine-tuning it to act as an interviewer based on job descriptions. This achievement showcases Celia's ability to provide realistic and tailored interview experiences.

What we learned

We learned how to collaborate effectively as a team during the hackathon, especially given the 36-hour time limit. Time management proved crucial, and active engagement in developing the app throughout the period was essential. Additionally, we enhanced our technical skills in full-stack web development and learned how to fine-tune AI models, contributing significantly to our project's success.

What's next for Ai interview with Celia

Our next steps for Celia involve continuing to refine and enhance its capabilities to reach production-level quality. We aim to improve the AI's accuracy and responsiveness, ensuring a seamless and realistic interview experience. Additionally, we plan to expand the range of job roles and industries that Celia can handle, making it a versatile tool for job seekers. We will also focus on optimizing the user interface and backend performance to support a larger user base. Ultimately, our goal is to deploy Celia as a fully functional and reliable interview preparation tool, accessible to users worldwide.