AI-DRIVEN VIRTUAL MOCK INTERVIEW

DEVELOPMENT
2024 Joint 13th International Conference on Soft Computing and Intelligent Systems and 25th International Symposium on Advanced Intelligent Systems (SCIS&ISIS) | 979-8-3503-7333-2/24/$31.00 ©2024 IEEE | DOI: 10.1109/SCISISIS61014.2024.10760210

Prabhat Kumar Mishra Arun Kumar Arulappan In-Ho Ra

School of Computer Science School of Computer Science School of Computer & Software
Engineering and Information Systems Engineering and Information Systems Kunsan National University
VIT University VIT University Gunsan, Republic of Korea
Vellore, India Vellore, India [email protected]
[email protected] [email protected]

Thanga Mariappan L Gina Rose G Young-Seok Lee

School of Computer Science Department of Computer Technology School of Computer & Software
Engineering and Information Systems Anna University Kunsan National University
VIT University Chennai, India Gunsan, Republic of Korea
Vellore, India [email protected] [email protected]
[email protected]

Abstract—The integration of Artificial Intelligence (AI) into aiming not only to enhance educational outcomes but also to
educational technologies marks a significant shift in learning make these sophisticated technologies accessible and
methodologies and operational dynamics within educational affordable for a diverse range of learners. The development of
institutions. At the forefront is an AI-driven virtual mock the company’s virtual mock interview platform stands as a
interview platform designed to address the high Customer cornerstone of this initiative. This platform leverages an
Acquisition Costs (CAC) in the edtech sector, especially for impressive array of AI tools, including OpenAI’s GPT-4, to
interview preparation services. This initiative harnesses a blend deliver personalized, context-aware training and feedback that
of AI technologies, including ADA 2 for creating context-aware mirrors real-life interview scenarios [2]. What sets this
embeddings and Machine Learning (ML), to transform the
platform apart is its commitment to ethical AI practices and
traditional mock interview process into a dynamic, cost-effective
system. Central to the platform is its use of advanced Natural
meticulous data management, poised to establish new
Language Processing (NLP) techniques and GPT-4 Large benchmarks for future technology deployment in education.
Language Model (LLM), automating the process of mock The challenges within the edtech sector, particularly the
interviews and providing personalized feedback, ensuring a prohibitive costs associated with traditional interview
preparation journey that meets specific candidate needs and preparation methods, are significant. These conventional
mirrors real interview scenarios. A key evaluation among 100 methods often lack the flexibility and personalization
students from a cohort of 1800 demonstrated a 90% cost necessary to meet the diverse needs of today’s learners and do
reduction for three mock interviews, reducing expenses from not fully leverage the potential of modern AI technologies.
₹3000 to just ₹300 per candidate. This cost efficiency The company identified these gaps as critical opportunities for
significantly enhances access to quality interview preparation, innovation.
improving student satisfaction and accessibility. Moreover, the
platform provides valuable insights into student performance, The institution envisioned a system that could transform
setting a new standard in educational technology by offering an the interview preparation landscape by employing AI to create
effective, personalized interview preparation experience. This a dynamic, scalable, and cost-effective solution. The goal was
project reflects a holistic approach to student development and to develop a platform that adapts to individual learning styles
the critical role of technology in addressing the evolving needs and needs while dramatically reducing operational costs. This
of learners approach not only improves the quality and accessibility of
interview preparation services but also enhances the overall
Keywords— Advanced Artificial Intelligence, ADA 2 user experience, lowering barriers to entry and broadening the
Embeddings, Machine Learning Innovations, Virtual Mock impact of educational innovations. The success of this
Interviews, Python Programming, Natural Language Processing, approach is evidenced by a substantial reduction in costs
Sentiment Analysis, CAC, Mock Interview. associated with mock interviews, which has been reduced by
90 percent, from ₹3000 to just ₹300 per candidate. This cost
I. INTRODUCTION
efficiency significantly broadens access to quality interview
The field of Educational Technology (EdTech) is preparation, greatly enhancing student satisfaction and
experiencing a profound transformation, driven by rapid educational equity [3]. Additionally, the platform equips the
advancements in Artificial Intelligence (AI) and Machine company with invaluable insights into student performance,
Learning (ML). This evolution is expanding opportunities helping educators tailor their approaches more effectively and
within the sector but also introduces significant challenges, setting a new standard for personalized educational tools.
particularly in specialized areas like interview preparation
where high Customer Acquisition Costs (CAC) pose a Mock interviews, an essential component of job
substantial barrier. Leading the innovation in tackling these preparation, offer candidates a platform to practice their
challenges is an unnamed edtech company, which has interviewing skills in a setting that mirrors real interview
embraced cutting-edge AI technologies to revolutionize conditions. However, the effectiveness of these sessions is
traditional educational methodologies [1]. This initiative often limited by several factors, including the inability to
integrates innovative tools such as ADA 2, Google Looker simulate the full intensity of a real interview and the quality
Studio, and advanced Natural Language Processing (NLP), and relevance of feedback provided [4]. Traditional mock

XXX-X-XXXX-XXXX-X/XX/$XX.00 © 20XX IEEE

Authorized licensed use limited to: VIT University. Downloaded on January 14,2025 at 14:38:58 UTC from IEEE Xplore. Restrictions apply.
interviews frequently fail to replicate the spontaneity and their prospective interviews. These systems cover a broad
stress of genuine interviews, leading to a false sense of range of scenarios and provide precise guidance to improve
preparedness among candidates [5]. Moreover, the feedback confidence, communication, and problem-solving skills [8].
from interviews often lacks depth and relevance, particularly
for specialized or niche roles, and the personal biases of Candidates using AI-enhanced interview platforms often
interviewers can skew the feedback, further complicating the report improved preparation and reduced anxiety, benefiting
training experience. To enhance the efficacy of mock from realistic practice and feedback. Studies show higher
interviews, several measures need to be implemented, success rates among candidates using these systems compared
including enhanced training for interviewers, expansion of to traditional methods. The structured feedback format makes
scenario ranges to cover more diverse job roles, and preparation efficient and effective, leading to better interview
advancement in AI technology to better simulate human-like performance and user satisfaction. Machine learning models
interactions. Additionally, developing structured support are vital in creating personalized learning experiences,
systems that help candidates apply feedback effectively analysing data to adjust content, pace, and complexity for each
through additional practice sessions, follow-up reviews, and student's needs [9]. This customization increases engagement,
targeted training modules will be crucial. This comprehensive making learning relevant and inclusive for all students.
Personalized learning optimizes understanding and aligns
approach aims to prepare candidates more effectively for the
challenges of real job interviews, addressing both the practical content with individual educational journeys. Machine
and emotional dynamics involved. learning also powers predictive analytics, allowing early
interventions by forecasting student performance based on
II. RELATED WORK data patterns. These insights help educators proactively
support students and refine teaching strategies. Predictive
AI is revolutionizing education with a host of innovative analytics improve learning outcomes by enabling tailored
applications that improve learning experiences. Adaptive support and personalized paths [10]. Quantum computing
learning systems analyse student performance, adjusting task accelerates data processing, creating adaptive learning
difficulty to individual abilities for a personalized challenge. platforms that provide immediate feedback and personalized
Intelligent tutoring systems offer one-on-one guidance, paths. This enables real-time content adaptation to individual
providing tailored feedback to address learning gaps. AI- learning speeds and styles.
driven educational games blend learning with game design,
engaging students through dynamic challenges that stimulate Augmented Reality (AR): AR overlays digital information
cognitive skills. Virtual Reality (VR) simulations immerse onto the real world, providing interactive, immersive learning
students in real-world environments, helping them grasp experiences. Students visualize complex concepts through 3D
complex concepts in fields like science and history. models, enhancing engagement and understanding.
Furthermore, AI tools like chat bots assist with student queries
Blockchain: Blockchain securely stores educational records
and administrative tasks, making education more efficient and
and credentials, preventing fraud and ensuring transparency.
accessible. AI profoundly transforms student learning and
This decentralized ledger provides tamper-proof storage,
assessment. Real-time analytics and adaptive feedback create
managing achievements efficiently.
personalized learning paths that address each student's unique
strengths and challenges. Predictive analytics, using vast Internet of Things (IoT): IoT devices in smart classrooms
amounts of data, forecast student performance and allow for range from attendance tracking to interactive smart boards.
early interventions to support struggling students. Automated They generate data streams that AI analyses for real-time
grading systems provide immediate feedback and maintain classroom optimization and personalized interactions.
objectivity, reducing bias while enhancing efficiency. AI
promotes formative assessments focused on skill development, Big Data Analytics: Big data aggregates and analyses vast
ensuring personalized learning and streamlined education data to uncover patterns related to performance and
processes. NLP technologies have significantly improved preferences. These insights help educators refine strategies
language comprehension for specialized applications like and improve outcomes.
interview simulations [6]. AI systems now understand Integrating AI with emerging technologies creates a
complex human language, interpret responses, and create dynamic educational landscape, where teaching and learning
human-like interactions for lifelike interview training. seamlessly blend with technology for optimized, personalized,
Context understanding, emotional tone, and conversational and secure learning experiences..
flow are key features that mimic real-world scenarios. This
technology helps job seekers practice responding to questions III. PROBLEM STATEMENT
and provides actionable feedback to refine both verbal and In the rapidly growing field EdTech, high customer
non-verbal skills. In educational and training platforms, NLP acquisition costs, particularly in interview preparation
plays a pivotal role in automated feedback systems. These services, are a significant challenge for service providers and
tools assess written and spoken submissions, offering detailed learners alike. Traditional mock interview methods tend to be
critiques on grammar, syntax, argument strength, and costly and rigid, lacking the flexibility and personalization
coherence [7]. They also include semantic analysis to required to meet diverse learner needs. They also do not fully
understand meaning and intent, providing comprehensive harness the potential of AI, which can provide insightful data
feedback for language learners. This approach democratizes on candidates' performance and preparation needs. One
education by delivering personalized, high-quality insights company identified these shortcomings as an opportunity to
that traditionally required human intervention. AI-enhanced innovate and transform interview preparation using AI. Their
mock interviews are an advanced training tool, analysing vision was to create a scalable and cost-effective platform that
speech patterns, relevance, and style to deliver impartial adapts to individual learning styles and needs while reducing
feedback. Tailored questions focus on specific job roles, operational costs. This platform aimed to enhance the quality
helping candidates experience a training session simulating and accessibility of interview preparation services while

Authorized licensed use limited to: VIT University. Downloaded on January 14,2025 at 14:38:58 UTC from IEEE Xplore. Restrictions apply.
improving the overall user experience, lowering barriers to adherence, and features for easy data access, export, and
entry, and broadening EdTech’s impact. deletion. Regular backups and a robust disaster recovery plan
ensure data security and recoverability.
The success of this approach is evident in the dramatic
reduction of mock interview costs, which have been cut by 90
percent from ₹3000 to ₹300 per candidate. This cost efficiency
significantly improves access to quality interview preparation,
increasing student satisfaction and promoting educational
equity. Additionally, the platform provides the company with
crucial insights into student performance, helping educators
refine their teaching strategies and set a new standard for
personalized educational tools.
IV. TECHNOLOGICAL FRAMEWORK
Our system harnesses advanced AI technologies to
revolutionize mock interviews through speech recognition and
processing for voice-based interaction. At the heart of this
platform is GPT-4, an advanced generative model developed
by OpenAI, capable of generating realistic and contextually
appropriate questions and responses. GPT-4, known for its
deep learning capabilities based on transformer architectures, Fig. 1. System Architecture
understands and generates human-like text with high fluency
and relevance by considering the entirety of a conversation or VI. RAG - RETRIEVAL AUGMENTED GENERATION
interview context. This ensures a natural interview flow with Retrieval-Augmented Generation (RAG) is an innovative
tailored questions. Additionally, GPT-4 can parse complex hybrid methodology in artificial intelligence, particularly
sentence structures and industry-specific jargon, making it within NLP, combining information retrieval systems and
suitable for a wide range of mock interviews across different neural network-based generation models. RAG uses external
fields. The platform employs machine learning algorithms to databases to enrich contextually relevant responses. Retrieval
analyse user responses during mock interviews, evaluating systems search large databases using techniques from
factors like response length, relevance, keyword inclusion, keyword-based methods to advanced semantic searches.
and hesitation to assess user performance comprehensively. Neural network generators, like GPT or BERT, produce
Utilizing historical data and ongoing performance metrics, our coherent, contextually appropriate text. The RAG process
system's predictive modelling identifies patterns in user starts with a user query, retrieves relevant data, and integrates
responses, indicating areas of weakness or strength. Based on it for the neural generator to produce a comprehensive
these insights, the platform adapts future sessions to focus on response. Key components include the retrieval system, neural
weaker areas, providing personalized and effective interview model, and diverse knowledge base. Some RAG systems use
practice sessions. This adaptive learning approach helps users reinforcement learning to refine processes. Applications
improve rapidly and gain confidence in their abilities. include enhancing question-answering systems, content
Seamless voice interaction is crucial for simulating real-life creation, journalism summarizing, and advanced educational
scenarios, supported by state-of-the-art speech-to-text tools. RAG offers improved response accuracy, access to
technologies that accurately convert spoken words into text. current data, and versatility across domains like customer
Real-time processing allows the AI to evaluate spoken support, academic research, healthcare, and legal assistance.
responses instantly, enhancing learning through immediate This methodology is particularly beneficial for fields requiring
feedback. Advanced algorithms ensure accurate transcription, detailed and precise information, ensuring users receive the
even with various accents or slightly noisy environments. most relevant and up-to-date content available.
Together, these AI technologies form a robust framework
powering the mock interview platform, leveraging GPT-4 for
realistic dialogues and comprehensive performance analysis
to enhance user preparation and success.
V. SYSTEM ARCHITECTURE AND DESIGN
The platform provides a seamless experience on web and
mobile with a responsive design for consistent performance
across devices. Special attention is given to UX design,
ensuring intuitive navigation with clear labelling, consistent
layout, and easy-to-use interactive elements. Accessibility is a
key focus, supporting screen readers and keyboard
navigability for users with disabilities. The back end uses
scalable cloud infrastructure (AWS, Azure, Google Cloud) for
flexibility, crucial during high usage periods like job fairs. A
microservices architecture allows independent scaling of
components like user management and data processing, with
Fig. 2. RAGs Architecture
event-driven architecture enabling real-time data processing
and instant feedback during interviews. Data handling is
secure and compliant with encryption, GDPR/HIPAA

Authorized licensed use limited to: VIT University. Downloaded on January 14,2025 at 14:38:58 UTC from IEEE Xplore. Restrictions apply.
 VII. IMPLEMENTATION AND RESULT ANALYSIS
Students record their answers, rated on correctness and
Evaluation and Implementation Strategy for AI-Driven
completeness. The graph compares five candidates, with
Mock Interview Platform. The development of an AI-driven
higher scores indicating better answer quality. Scores are
mock interview platform requires a detailed evaluation
directly proportional to the quality of the candidates'
framework and a strategic implementation plan. By
responses.
integrating advanced AI technologies with user-centric
design, this platform aims to revolutionize the way users
prepare for interviews. Here is an outline of the key
evaluation metrics, technical foundations, and strategic
milestones essential for the successful rollout of the platform,
complemented by relevant UML diagrams to provide visual
context and clarity.

Fig. 5. Comparison of Candidate Performance based on answers

correctness and completeness

ACKNOWLEDGMENT
This research was supported by "Regional Innovation
Strategy (RIS)" through the National Research Foundation of
Korea (NRF) funded by the Ministry of Education
(MOE)(2023RIS-008).
Fig. 3. Use Case Diagram REFERENCES
[1] Feng, S. Y., Gangal, V., Wei, J., Chandar, S., Vosoughi, S., Mitamura,
Evaluation Metrics for AI-Driven Platform: T., & Hovy, E. (2021). A survey of data augmentation approaches for
Accuracy of AI Responses: Aim for 85% accuracy initially, NLP. arXiv preprint arXiv:2105.03075.
increasing to 95% post-optimization, by comparing AI- [2] Kiela, D., Bartolo, M., Nie, Y., Kaushik, D., Geiger, A., Wu, Z., ... &
generated answers against a gold standard set by industry Williams, A. (2021). Dynabench: Rethinking benchmarking in NLP.
arXiv preprint arXiv:2104.14337.
experts. User Success Rate: Track an average 20%
[3] Verrap, R., Nirjhar, E., Nenkova, A., & Chaspari, T. (2022, December).
improvement in user interview skills across ten sessions. “Am I Answering My Job Interview Questions Right?”: A NLP
System Response Time: Maintain an average response time Approach to Predict Degree of Explanation in Job Interview Responses.
under 2 seconds to ensure user engagement. Session In Proceedings of the Second Workshop on NLP for Positive Impact
(NLP4PI) (pp. 122-129).
Reliability: Achieve 99.9% system uptime during active user
[4] Li, M., Chen, X., Liao, W., Song, Y., Zhang, T., Zhao, D., & Yan, R.
sessions. (2023, February). EZInterviewer: To Improve Job Interview
Scalability and Security Tests: Load Testing: Support up Performance with Mock Interview Generator. In Proceedings of the
to 10,000 concurrent users without performance degradation. Sixteenth ACM International Conference on Web Search and Data
Stress Testing: Ensure the system maintains functionality at Mining (pp. 1102-1110).
150% of normal load. Security Auditing: Adhere to ISO [5] Tolle, H., Castro, M. D. M., Wachinger, J., Putri, A. Z., Kempf, D.,
Denkinger, C. M., & McMahon, S. A. (2024). From voice to ink
27001 and GDPR, targeting zero breaches. (Vink): development and assessment of an automated, free-of-charge
transcription tool. BMC Research Notes, 17(1), 1-11.
[6] Ranzenberger, T., Bocklet, T., Freisinger, S., Georges, M., Glocker,
K., Herygers, A., ... & Zakaria, K. (2024). Extending HAnS: Large
Language Models For Question Answering, Summarization, And
Topic Segmentation In An ML-based Learning Experience Platform.
In Elektronische Sprachsignalverarbeitung 2024, Tagungsband der
35. Konferenz, Regensburg, 6.-8. März 2024 (pp. 219-224).
TUDpress.
[7] Zhang, X., Yu, B., Yu, H., Lv, Y., Liu, T., Huang, F., ... & Li, Y.
(2023). Wider and deeper llm networks are fairer llm evaluators.
arXiv preprint arXiv:2308.01862.
[8] Malam, V. K. Natural Language Processing-based Solution for
Accurate Transcription and Translation of Distorted Multilingual
Audio Signals.
[9] Malam, V. K. Natural Language Processing-based Solution for
Accurate Transcription and Translation of Distorted Multilingual
Audio Signals
[10] Xian, J., Teofili, T., Pradeep, R., & Lin, J. (2024, March). Vector
search with OpenAI embeddings: Lucene is all you need. In
Fig. 4. Activity Diagram Proceedings of the 17th ACM International Conference on Web
Search and Data Mining (pp. 1090-1093).

Authorized licensed use limited to: VIT University. Downloaded on January 14,2025 at 14:38:58 UTC from IEEE Xplore. Restrictions apply.