Cerebras and AIBI Transform AI Interviewing Methods

In the fast-paced world of recruitment, where finding the right talent is more crucial than ever, a groundbreaking innovation has emerged to streamline the hiring process. Introducing AIBI, the first comprehensive AI-driven interview bot that promises a seamless, real-time interview experience. AIBI is designed to conduct realistic interviews, generating high-quality, real-time technical questions and responses without any noticeable delay. This is made possible through the use of Cerebras inference, which has significantly enhanced its speed and efficiency, reducing both the "time to first token" and the token generation speed by an impressive 66-80%.

A Comprehensive End-to-End Solution

AIBI serves as an end-to-end recruitment solution, efficiently managing various stages of the hiring process. Here’s how it works:

1. Generating Initial Job Applications: AIBI begins the process by creating a Google Form for job applications, ensuring that the initial step is both straightforward and comprehensive.
2. Creating Tailored Interview Questions: It crafts candidate-specific interview questions, following a logical and responsive flow, to ensure that each interview is unique and pertinent to the applicant’s profile.
3. Conducting Real-Time Audio Interviews: The bot carries out the actual interviews, engaging candidates in real-time conversations that mimic the dynamics of human interaction.
4. Generating Reports and Evaluating Candidates: After the interview, AIBI produces a detailed report, evaluating the candidate’s performance and providing insights for the recruitment team.
   

   Addressing Key Challenges in Recruitment
   

   The recruitment process is filled with challenges such as the need for speed, personalization, and responsiveness. Hiring the right talent is one of the most critical tasks for any organization, and it’s a time-consuming endeavor. As emphasized by OpenAI CEO, Sam Altman, leaders should spend a significant portion of their time on hiring. However, traditional interview methods often fall short in terms of scalability, consistency, and personalization. AI-powered interview bots have been around, but they frequently struggle with real-time personalization and responsiveness, which are crucial for maintaining candidate engagement. Slow inference speeds can disrupt the interview flow, leading to a frustrating experience and potential candidate dropout. Moreover, relying on static, pre-generated questions limits the ability to evaluate candidates thoroughly.

   The Evolution of AIBI
   

   AIBI stands out as an innovative AI recruiter that manages the entire recruitment process. Initially powered by GPT-4o, AIBI transitioned to using Llama70B on Cerebras, achieving a remarkable 72% reduction in latency and significantly enhancing the candidate experience. With earlier models like GPT-4o, candidates encountered delays of several seconds for each generated response, which were further prolonged when converting these responses into speech. Follow-up questions could experience up to 10 seconds of delay per interaction. However, with Cerebras inference, AIBI has achieved an average latency reduction of 75%.

   The Architecture Behind AIBI
   

   AIBI’s architecture is divided into two primary components: the HR Dashboard Platform and the AI Interviewer Bot.

   The HR Dashboard Platform
   

   The HR Dashboard Platform is responsible for generating customized job application forms, preliminary interview questions, and post-interview grading reports. Here’s how it functions:

   • Job Application Form Generation: The job description is input into a language model (LLM), which outputs a JSON containing 15-30 questions. This JSON is then directly parsed into a Google Form. With the transition from GPT-4o to LlaMa 3.1-70B on Cerebras, this process time was reduced from 4 seconds to less than 1 second, marking a 75% reduction in latency.
   • Preliminary Questions Creation: Once an applicant completes the Google Form, their responses are processed by the LLM to generate a unique set of preliminary questions tailored to the applicant. Switching from GPT-4o to LlaMa 3.1-70B on Cerebras reduced the time for this step from 3 seconds to under 1 second, a 66% improvement.
   • Evaluation Rubric Generation: The platform also creates an evaluation rubric specific to the role based on job requirements. This transition to LlaMa 3.1-70B resulted in a 75% latency reduction.

     Overall, the HR Dashboard component achieved an average latency reduction of 72%.

     The AI Interviewer Bot
     

     The AI Interviewer Bot is responsible for conducting the audio interview with the candidate. Once a candidate joins the interview, they are greeted with a pre-recorded message that explains the interview rules and requests their consent. Upon receiving consent, the bot begins asking uniquely generated questions.

     Asking Questions
     

     For each question generated by the LlaMA-70B running on Cerebras, it is converted into an audio file using OpenAI’s text-to-speech model, ‘tts-1’. This audio file is then sent to the front end and played to the candidate, who has up to two minutes to respond.

     Candidate Response
     

     As the candidate responds, their audio is transcribed to text using OpenAI’s speech-to-text model, ‘whisper-1’. The transcribed text is then analyzed by the LLM to determine if it contains a question or an answer. If it’s a question, the transcribed conversation is used as context to generate an answer. Otherwise, the AI Interview Bot proceeds with the next question. Before adopting Cerebras Inference, this series of calls to the LLM resulted in a slow and frustrating experience for candidates. It took 1-2 seconds for GPT-4o to determine if a response contained a question, followed by an additional ~3 seconds to generate a response. Each of these steps was reduced to less than 1 second with Cerebras.

     Overall, the AI Interviewer Bot achieved an average latency reduction of 75%.

     Future Prospects
     

     AIBI is just at the beginning of its journey and aims to continue evolving into an efficient, realistic, and intelligent interviewing tool. Some promising areas for future development include:

   • Incorporating AI-Generated Video: By adding video capabilities, the interview bot could simulate a face-to-face interview experience, making it more personal and engaging.
   • Enhancing Multi-Round Interviews: Expanding the system’s capacity to handle multi-round interviews and follow-ups could further streamline complex recruitment processes.
   • Implementing Advanced Agents: Using more sophisticated agents to adjust the interview based on candidate responses could allow for dynamic interview paths, making the process more adaptable and insightful.

     For those interested in contributing to this innovative project or suggesting new features, you can explore the code on GitHub: AIBI Project on GitHub.

     In conclusion, AIBI represents a significant leap forward in recruitment technology, offering a more efficient and engaging way to conduct interviews. By leveraging advanced AI and reducing latency, it addresses the key challenges of personalization and responsiveness in the hiring process, paving the way for a more dynamic and effective recruitment landscape.

For more Information, Refer to this article.