An AI-Based Virtual Interview Assistant for Adaptive Interview Simulation and Feedback using NLP

An AI-Based Virtual Interview Assistant for Adaptive Interview Simulation and Feedback using NLP

Richa Saxena

Department of Computer Science Moradabad Institute of Technology

Moradabad, India richasaxena2006@gmail.com

Shubham Bhatt Department of Computer Science Moradabad Institute of Technology

Moradabad, India bhattsibbu@gmail.com

Ansh Raj Saxena Department of Computer Science Moradabad Institute of Technology

Moradabad, India saxenaanshraj02@gmail.com

Aarohi Saxena Department of Computer Science Moradabad Institute of Technology

Moradabad, India aarohisaxena1014@gmail.com

Aarohi

Department of Computer Science Moradabad Institute of Technology Moradabad, India aarohisingh943@gmail.com

AbstractInterview preparation plays a vital role in determin- ing a candidates employability; however, access to personalized and scalable interview practice remains a signicant challenge for many students and job seekers. Conventional preparation methods such as manual mock interviews, coaching sessions, and static question banks are either resource-intensive or insuf- ciently adaptive to individual candidate proles. As a result, many candidates face difculty in articulating their knowledge and experience effectively during real interviews. This paper presents a Virtual Interview Assistant designed to simulate personalized interview experiences using resume-based question generation and automated feedback mechanisms. The proposed system accepts a candidates resume as input and extracts relevant information such as skills, education, work experience, and projects using natural language processing techniques aligned with Applicant Tracking System (ATS) standards. Based on this extracted information, the system dynamically generates technical and human resource interview questions tailored to the candidates background. During the virtual interview session, candidates respond to questions through voice-based interaction. The system evaluates these responses using language under- standing techniques to assess relevance, clarity, and conceptual correctness. Structured feedback is then provided to highlight strengths, identify gaps, and suggest areas for improvement. By integrating resume analysis, adaptive question generation, interview simulation, and feedback evaluation into a unied framework, the Virtual Interview Assistant aims to deliver a realistic and scalable interview preparation environment. The proposed framework is application-oriented and product-focused, making it suitable for deployment in academic institutions, placement training centers, and self-paced learning platforms. The system demonstrates how AI-driven interview simulation can enhance candidate readiness, condence, and overall interview performance.

Index TermsVirtual Interview Assistant, Resume-Based Question Generation, Natural Language Processing, Applicant Tracking System (ATS), Voice-Based Interview Simulation, Au- tomated Feedback, Interview Preparation, Articial Intelligence.

1. INTRODUCTION

   The rapid growth of the information technology sector, the growing pressure in employment circles has focusing more on how crucial it is to get ready properly for interviews [15]. Though schools emphasize theory alongside hands-on tasks, Some folks know a lot, yet turning that into success trips up plenty of learners and those hunting work turning what they know into clear answers that hold together well when asked questions.

   Finding a job sometimes means showing skill on paper – yet what happens in the room matters just as much Clear com- munication matters. What counts is how well ideas connect, thinking straight helps too. The way thoughts link makes a difference because they can talk about what happened before in a clear way, things often work out better Fumbling through questions often trips people up when they face job interviews. Not knowing what to expect slows many down right at the start. Missing practice means stumbling under pressure later on Getting a job remains tough, especially if you have technical skills but face stiff competition.

   In recent years, advancements in articial intelligence and Computers understanding human speech has led to new tools being built Some smart setups can grasp whats being said while also producing responses that make sense. These tools learn patterns without needing every detail spelled out ahead of time Lots of folks now work with machines that understand how people talk. Such tools are used everywhere these days used now in things like automated helpers, tools that suggest what you might like, auto-marking tools plus digital helpers. [1], [2], [3].

   Whatstill, using them to get ready for job talks hasnt caught on much still stuck using basic quiz collections or

   scripted practice tests interview platforms that do not adapt to individual candidate proles. A resume represents a structured summary of a candidate. A persons schooling shows their learning path. What they can do comes through abilities gained over time. Work done highlights real tasks completed. Past jobs give a view of growth step by step. Few tools out there manage what they promise preparing you properly even though it matters so much make good use of whats on your resume. Questions in interviews often come from it usually xed, bearing little connection to what the person really brings from their past practice, and it feels less real, so it helps less when you need it most. Furthermore, feedback provided by existing platforms is typical, oddly shallow, caring more about how long answers are than what they mean.

   Instead of things like how well it ts or makes sense. This study tackles those issues through a suggested approach Virtual Interview Assistant powered by resume data using insights to shape how interviews feel for each person. Looking at patterns helps adjust the approach to the candidates resume and generating customized interview with questions in play, realism grows through how the system behaves internally. On top of that, machines can grade answers without help. Still, setting up the space takes time before anyone walks in mechanisms for feedback help steer applicants along the way, and improvement is possible. This approach shows how things could work differently and how AI-driven systems can enhance interview preparation. It shifts as needs change, grows when required, yet stays within reach.

2. LITERATURE REVIEW

   Nowhere has change been clearer than in how smart tools help people learn and choose work paths. Machines that un- derstand speech and text shape what comes next in education guidance. These shifts started quietly but now affect choices in real ways. Learning systems think differently because computers read more like humans do. Progress here didnt come fast yet it reshapes advice given today. A key part of current studies looks at how machines can judge student work while shaping lessons to t individual needs. In these setups, software watches what users do, then adjusts responses based on that behavior. Work in this space has appeared often across different elds take smart tutors, computer scoring, or chat helperseach shows how language processing can assist real teaching and testing work.

   Looking into hiring and job readiness, researchers have explored how computers can pull skills from resumes using language tools. Instead of just scanning words, these systems spot key details through pattern detection or smart algorithms. Some setups go a step beyond by comparing applicant back- grounds to role needs automatically. While such tech speeds up sorting candidates, it stumbles when faced with varied layouts or unclear skill contexts. How well it works often depends on how messy real-world resmes tend to be.

   A fresh look at automation turns toward creating questions by machine. Earlier work dipped into pulling queries from texts like manuals, journals, papers, or stored factsusing

   sentence structure analysis, meaning detection in phrases, also newer deep learning networks built on transformers. Some of these methods do well when tailored to certain topics, tting classroom needs with decent accuracy. Yet trying them out where job seekers practice interviews, drawing prompts straight from a resume, hasnt drawn much attention so far.

   One key area looks at how machines give feedback and judge answers. Studies on grading brief replies or conver- sations show computers can check if ideas match, stay on topic, or make sense together. Even though such tools work well when tasks are xed, using them for free-form interview answers remains tough. So far, past work builds a solid base for AI help in prepping interviews – yet leaves room where new solutions could step in.

3. RELATED WORK

   Folks building software for job practice havent cracked the code yet. Some apps pair people together online, tossing out common questions one after another. Tools like Interview- Buddy or Pramp run drills that feel close to real chats. These sessions can help, sure, but only if someone else shows up or the script doesnt stale. Over time, it gets harder to grow without fresh angles or tailored feedback.

   Some studies built chatbot tools using xed scripts or rigid conversation rules. While they handle simple exchanges well, they cannot adjust to a persons unique background. Lately, smarter versions use AI trained on data to run interviews. Still, most pay more attention to managing talk ow than matching questions to someones work history.

   Fig. 1. Year-wise growth in research publications related to AI-based interview preparation and assessment systems.

   Looking at whats already out there, seen in Fig. 1, the new tool stands apartnot through one feature but how it brings together resume review, smart questioning, practice sessions, along with clear responses, all working under one system. A fresh angle emerges when these pieces connect, not stand apartshaping interviews that feel real and t each person. What sets this apart sits in the middle ground: part learning tool, part smart hiring aid, lling a space most tools miss.

   A few research efforts suggest tools powered by articial intelligence that sort applicants, matching resumes to job needs. While faster hiring comes out of this, the focus leans toward helping employers, not job seekers. These tools rarely highlight ways to get better at interviews or grow skills. Whats more, any response they give tends to be hidden in numbers or positions, missing clear reasons behind the results.

4. PROBLEM STATEMENT AND OBJECTIVES

   A doorway opens when theory meets practice interviews link classroom lessons to real jobs. Yet even those who know their subject well tend to fumble under pressure, held back by shaky nerves, poor readiness, and few chances to rehearse tough questions. Most prep routes today depend on live instructors or group drills, which eat up hours, money, and energy while leaving countless learners behind. Feedback, if offered at all, arrives late, vague, or one-size-ts-all rather than sharp, timely, built for growth.

   With the rise of online tools for interview preparation now exist, yet they often depend on xed sets of common questions delivered in rigid sequences. Most skip using a persons actual resume, so the practice questions rarely match what someone truly knows or has done. Instead of deep analysis, feedback tends to count words or spot keywords without judging how clear, relevant, or well-expressed an answer really is.

   A fresh approach tackles those issues head-onintroducing a Virtual Interview Assistant driven by natural language processing and smart algorithms. Instead of generic drills, it reads a persons resume carefully, pulling out relevant details to shape unique question sets. Questions come alive through individual proles, shaped by past roles, education, and skills. When answers arrive, the system dissects them with careful analysis rooted in how machines grasp human speech. Personalized replies meet organized insights, nudging users toward sharper performance under pressure. Condence grows quietly, built not on guesswork but repeated practice. Communication sharpens each round, rened through targeted input. Built for expansion, the tool ts learners, applicants, and career centers alike. This isnt just another test run. It adapts easily and is meant for real-world demands.

5. PROPOSED SYSTEM AND METHODOLOGY

   1. System Architecture

      The Virtual Interview Assistant is designed as a modular system in which each component performs a specic func- tion within the overall interview preparation pipeline. This modular architecture improves maintainability, scalability, and

      exibility [10], allowing individual modules to be enhanced or replaced without affecting the entire system.

      The Virtual Interview Assistant Fig. 2 follows a structured architecture that transforms a candidates resume into an interactive and personalized interview experience. The system is designed [11] as a layered pipeline in which data ows sequentially through interconnected modules, ensuring smooth processing and clear separation of responsibilities.

      Fig. 2. System Architecture

   2. Resume Input and Parsing

      The rst module handles resume input and preprocessing. Candidates upload their resumes in commonly used formats such as PDF. The system extracts textual content using docu- ment parsing libraries and applies natural language processing techniques to identify structured information such as personal details, education, skills, work experience, certications, and projects [7], [8], [9]. This extracted information forms the foundation for all subsequent modules. Special attention is given to handling variations in resume formats to ensure robustness and accuracy.

   3. ATS and Resource Analysis

      In this module, the parsed resume content is evaluated against industry-relevant keywords and Applicant Tracking System (ATS) standards [16]. The system identies missing or weak skill indicators by comparing resume data with predened job proles and role-based skill sets [17]. This analysis helps determine the candidates preparedness for specic roles and provides insight into gaps that may affect interview performance.

   4. Resume-based Question Generation

      Based on the extracted resume data and ATS analysis, the system dynamically generates interview questions [4]. These questions include technical, conceptual, and human resource- based queries tailored to the candidates skills, experience, and academic background [5]. Unlike static question banks, this module ensures contextual relevance, making the interview experience more realistic and personalized [6].

   5. Virtual Interview Simulation

      The virtual interview module simulates a structured inter- view environment. Candidates interact with the system through text or voice-based inputs, responding to generated questions in real time [2]. The interview ow is adaptive, allowing follow-up questions or difculty adjustments based on the candidates responses [5].

   6. Answer Evaluation and Feedback

   Candidate responses are evaluated using natural language understanding techniques to assess relevance, coherence, and conceptual accuracy. The system generates feedback high- lighting strengths, areas of improvement, and suggestions for enhancing responses.

   the mathematical formulation and implementation of the per- formance scoring system.

   1. Performance Score Denition

      Let us dene a performance score P for each mock inter- view session as follows:

      P = w1 · A + w2 · C + w3 · T w4 · E (1) where he parameters are dened in Table I.

      TABLE I PERFORMANCE SCORING PARAMETERS

      Symbol	Meaning
      P A C T E w1, w2, w3, w4	Overall Performance Score Accuracy % of questions answered correctly Condence Level measured from voice/text analysis (scaled 01) Timeliness average response speed (scaled 01) Errors penalty score for incorrect logical structure or off-topic responses Weights assigned based on importance of each metric

   2. Component Metrics

   Accuracy (A): This metric represents the proportion of interview questions answered correctly by the candidate. If a user answers 8 out of 10 questions correctly:

   8

   A = = 0.8 (2)

   10

   Fig. 3. End-to-end workow of the Virtual Interview Assistant from resume input to feedback generation.

   G. Job Role Recommendation

   Based on resume analysis and interview performance, the system recommends suitable job roles and career paths, assist- ing candidates in aligning their skills with market demands.

   The workow shown in Fig. 3illustrates the sequential processing pipeline of the proposed system, highlighting how resume data is transformed into personalized interview ques- tions and actionable feedback through AI-driven analysis.

6. SCORING METHODOLOGY

   To provide quantitative feedback and measurable progress tracking, the Virtual Interview Assistant implements a com- prehensive scoring mechanism that evaluates candidate per- formance across multiple dimensions. This section presents

   Higher values indicate better technical correctness and do- main knowledge [18]. Accuracy is calculated by comparing candidate responses against expected answer patterns using natural language understanding techniques.

   Condence (C): This metric captures the non-verbal ef- fectiveness and clarity of responses. Condence is mea- sured through speech analysis, text tone evaluation, or self- assessment ratings:

   C [0, 1] (3)

   For instance, if acoustic or textual analysis identies mod- erate condence in the candidates responses:

   C = 0.6 (4)

   This component evaluates how clearly and condently can- didates express their knowledge during the interview.

   Timeliness (T ): This metric measures response promptness and efciency. It reects how quickly candidates formulate and deliver their answers relative to expected response times:

   T [0, 1] (5)

   Faster and more concise answers without compromising quality receive higher timeliness scores. This metric helps

   evaluate a candidates ability to think and respond under time pressure, which is crucial in real interview scenarios.

   Errors (E): This penalty component counts critical mis- takes such as off-topic responses, logical fallacies, or concep- tually incorrect statements. For example, if a candidate makes 2 major errors during the interview:

   E = 2 (6)

   The error count reduces the overall performance score, ensuring that quality is prioritized over other metrics.

   1. Weight Assignment

      The weights w1, w2, w3, w4 are assigned based on the relative importance of each metric in overall interview perfor-

      mance evaluation. A typical weight conguration might be:

      w1 = 0.4, w2 = 0.3, w3 = 0.2, w4 = 0.1 (7)

      These weights sum to 1.0 for normalized scoring, with

      accuracy receiving the highest weight as it directly reects the correctness of responses, followed by condence, timeliness, and error penalty respectively. The weight distribution can be adjusted based on specic interview types or organizational requirements.

   2. Example Calculation

      Consider a mock interview session where a candidate achieves the following metrics:

      • Accuracy: A = 0.8 (answered 80% of questions cor- rectly)

      • Condence: C = 0.7 (demonstrated good condence level)

      • Timeliness: T = 0.9 (responded promptly and efciently)

      • Errors: E = 1 (made one major error)

        Using the weight conguration w1 = 0.4, w2 = 0.3, w3 = 0.2, w4 = 0.1, the performance score is calculated as:

        P = 0.4(0.8) + 0.3(0.7) + 0.2(0.9) 0.1(1)

        Along with the numerical score, the system provides de- tailed feedback breaking down performance across individual metrics, helping candidates understand their strengths and areas needing attention. This quantitative approach enables systematic tracking of progress across multiple interview sessions and facilitates data-driven improvement in interview preparation.

7. RESULTS AND DISCUSSION

   Evaluation of the Virtual Interview Assistant was conducted by observing its performance and user interaction outcomes during live use of the deployed system. The platform success- fully demonstrates in Fig. 4 the seamless integration of resume parsing, adaptive question generation, interactive voice-based simulation, and automated feedback generation. Users reported that the system consistently generated interview questions that were relevant to their skills and experiences as specied in their resumes a key advantage over generic question banks.

   Fig. 4. Keywords Extraction from the uploaded Resume.

   The resume analysis component effectively captured struc- tured information from uploaded resumes and inuenced ques- tion relevance, conrming the utility of ATS-aligned parsing in interview preparation contexts. The voice-based interview simulation provided a realistic interface for users to practice responses under conditions closer to real interviews, which many users rated as more engaging, and condence-boosting

   = 0.32 + 0.21 + 0.18 0.10

   = 0.61

   (8)

   compared to text-only methods.

   Feedback delivered by the system highlighted areas such as response clarity, technical coverage, and communication

   The resulting performance score is 0.61 (out of 1.0), providing a quantitative measure of interview performance. This score can be tracked over multiple sessions to monitor candidate improvement and identify specic areas requiring additional practice.

   1. Score Interpretation and Feedback

      The performance score P is interpreted as follows:

      • P 0.8: Excellent performance

      • 0.6 P < 0.8: Good performance with room for improvement

      • 0.4 P < 0.6: Average performance requiring focused practice

      • P < 0.4: Requires signicant improvement

   effectiveness. Users noted that the feedback was constructive and helped them understand specic areas of improvement. However, some limitations were observed, particularly in accurately scoring highly varied linguistic expressions and in differentiating between nuanced correctness levels in open- ended responses. The absence of deeper semantic scoring and emotion recognition limits the systems ability to fully simulate human evaluation criteria. Fig. 5 presents a graph- based comparison of the proposed Virtual Interview Assistant with existing interview preparation platforms. Generic mock platforms and peer-based tools demonstrate limited personal- ization and feedback quality. Chatbot-based systems improve automation but still lack resume-driven contextual relevance. In contrast, the proposed system consistently achieves higher

   Fig. 5. Comparison of interview preparation platforms across key evaluation metrics.

   effectiveness scores across all metrics, particularly in question relevance, personalization, and interview realism. This high- ligts the impact of integrating resume analysis, ATS-based evaluation, and adaptive interview simulation within a unied framework.

   Overall, the results indicate that the Virtual Interview Assis- tant signicantly enhances user preparedness and engagement. Its combination of personalized question generation with adap- tive feedback provides a more meaningful interview practice experience than basic automated systems. Future work could incorporate quantitative user studies and performance metrics to further validate efciency and reliability.

   Similar trends in the adoption of AI-based interview and recruitment systems have been reported in recent academic surveys, indicating growing research interest in intelligent employability solutions [16].

8. TECHNOLOGY, TOOLS AND REQUIREMENTS

   The Virtual Interview Assistant is implemented using a combination of modern web technologies like Python [12], Next.js [13], natural language processing libraries, Hugging Face Transformers [14] and machine learning frameworks to ensure efciency, scalability, and ease of deployment. The selection of tools is guided by practical considerations such as open-source availability, community support, and compati- bility with real-world deployment environments.

   1. Software Requirements

      The frontend of the system is developed using web tech- nologies such as HTML, CSS, and JavaScript, with Next.js employed to build a responsive and interactive user interface. This framework enables efcient client-server communication and supports dynamic rendering of interview sessions and feedback dashboards.

      The backend is implemented using Python-based services integrated with the web framework. Natural language pro- cessing tasks such as resume parsing, keyword extraction, and response analysis are handled using libraries including NLTK, SpaCy, and transformer-based models where appropriate. Re- sume les are processed using document parsing tools such

      as PDFMiner, ensuring reliable text extraction from uploaded resumes.

      For data storage, relational or NoSQL databases such as MySQL or MongoDB are used to store structured resume data, interview logs, feedback reports, and user proles. Version control and collaboration are managed through GitHub, while deployment and containerization are supported using Docker and cloud platforms such as Vercel.

   2. Hardware Requirements

   The system is designed to run efciently on standard computing hardware. A personal computer or laptop with a minimum of 8 GB RAM, a modern processor, and stable internet connectivity is sufcient for development and testing. For large-scale deployment or voice-based processing, cloud servers with higher computational resources may be utilized.

9. FEASIBILITY STUDY AND IMPLEMENTATION CONSIDERATION

   Before deploying an intelligent interview preparation sys- tem, it is essential to evaluate its feasibility from economic, technical, and operational perspectives. The proposed Virtual Interview Assistant has been designed with practicality and scalability in mind, ensuring that it can be adopted in academic and professional environments without excessive resource re- quirements.

   1. Economic Feasibility

      The system is economically feasible as it relies primarily on open-source technologies and widely available development frameworks. Libraries used for natural language processing, resume parsing, and web development are freely accessible, re- ducing licensing costs. Additionally, the modular design allows institutions to deploy only required components, minimizing infrastructure expenses. Cloud-based deployment options fur- ther reduce the need for high-end local hardware, making the solution affordable for educational institutions and training centers.

   2. Technical Feasibility

      From a technical point of view, the proposed system utilizes mature and well-supported technologies. Natural language processing frameworks such as SpaCy and NLTK are ca- pable of handling resume analysis and text evaluation tasks efciently. Web technologies like Next.js enable the devel- opment of responsive and scalable interfaces, while backend services manage data processing and system coordination. The architecture supports both local and cloud-based deployment, ensuring exibility and reliability.

   3. Challenges and Constraints

   Despite its feasibility, the system faces certain challenges. Accurately evaluating open-ended interview responses remains complex, as human language can vary signicantly in structure and expression. Voice-based interaction introduces additional challenges related to speech recognition accuracy and back- ground noise. Moreover, maintaining fairness and reducing

   bias in automated evaluation requires careful model design and continuous renement. Addressing these challenges is essential for ensuring reliable and ethical system performance.

10. FUTURE SCOPE

    The proposed Virtual Interview Assistant provides a strong foundation for intelligent and personalized interview prepa- ration; however, several enhancements can further improve its effectiveness and applicability. One potential extension is the incorporation of advanced speech analysis techniques to evaluate voice modulation, uency, and condence levels during interviews. This would enable a more comprehensive assessment of communication skills beyond textual evaluation. Another promising direction is the integration of emotion and sentiment analysis to better understand candidate behavior under interview conditions. Detecting stress, hesitation or condence levels could help provide deeper feedback and simulate real interview pressure more accurately. Multilingual support can also be introduced to help candidates from diverse

    linguistic backgrounds and improve accessibility.

    Future versions of the system may include adaptive dif- culty adjustment, where the complexity of questions evolves based on real-time performance. In addition, integrating industry-specic interview patterns and organization-level cus- tomization can make the platform more suitable for targeted recruitment and role-specic preparation. Long-term analytics and progress tracking can help users monitor improvement over multiple interview sessions.

    Further research may also focus on reducing evaluation bias and improving fairness by rening NLP models with diverse datasets. With these enhancements, the Virtual Interview As- sistant can evolve into a comprehensive career preparation platform that supports interview readiness, skill development, and informed career decision-making.

11. CONCLUSION

This paper proposes a comprehensive framework for a Vir- tual Interview Assistant that utilizes articial intelligence and natural language processing to provide personalized, resume- driven interview preparation and automated feedback. The system is designed to overcome the limitations of traditional interview training methods by offering a scalable, adaptive, and accessible solution tailored to individual candidate pro- les. By employing NLP and ATS-based resume analysis, the assistant generates context-aware interview questions that closely align with real-world recruitment practices.

The proposed system follows a modular architecture inte- grating key components such as resume parsing, intelligent question generation, virtual interview simulation, response evaluation, and structured feedback delivery. This unied workow enhances exibility, maintainability, and scalabil- ity, making the framework suitable for deployment across academic institutions and professional training environments. Unlike conventional interview preparation tools, the system emphasizes adaptive learning and continuous performance improvement through data-driven feedback.

From an application-oriented perspective, the Virtual In- terview Assistant functions as a product-focused solution supporting placement training, self-paced interviw practice, and preliminary candidate assessment. By simulating realistic interview scenarios, the system helps candidates rene com- munication skills, build condence, and improve interview readiness within a low-pressure learning environment.

Although the current work is primarily conceptual and does not include extensive experimental validation, it establishes a strong and scalable foundation for real-world implementation. The proposed architecture demonstrates practical feasibility and adaptability, enabling future prototyping, pilot deploy- ment, and commercialization.

Future enhancements may include advanced speech ana- lytics for evaluating uency and tone, emotion-aware feed- back mechanisms to assess condence and stress levels, and multilingual support to improve accessibility for diverse user groups. Overall, the framework highlights the potential of AI- driven interview simulation to enhance candidate performance and align interview preparation with modern, technology- driven recruitment practices.

REFERENCES

1. S. Russell and P. Norvig, Articial Intelligence: A Modern Approach, 4th ed. Pearson, 2021.

2. D. Jurafsky and J. H. Martin, Speech and Language Processing, 3rd ed. (Draft). Prentice Hall, 2023.

3. I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT Press, 2016.

4. T. Mikolov, K. Chen, G. Corrado, and J. Dean, Efcient Estimation of Word Representations in Vector Space, Proc. Int. Conf. on Learning Representations (ICLR), 2013.

5. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, Proc. NAACL-HLT, 2019.

6. A. Vaswani et al., Attention Is All You Need, Proc. Advances in Neural Information Processing Systems (NeurIPS), 2017.

7. C. D. Manning et al., The Stanford CoreNLP Natural Language Processing Toolkit, Proc. ACL System Demonstrations, 2014.

8. M. Honnibal and I. Montani, spaCy 2: Natural Language Understanding with Bloom Embeddings, To appear, 2017.

9. S. Bird, E. Klein, and E. Loper, Natural Language Processing with Python. OReilly Media, 2009.

10. R. T. Fielding, Architectural Styles and the Design of Network-based Software Architectures, Ph.D. dissertation, Univ. of California, Irvine, 2000.

11. M. Masse, REST API Design Rulebook. OReilly Media, 2011.

12. Python Software Foundation, Python Language Reference, Version

    3.11. [Online]. Available: https://www.python.org/

13. Next.js Documentation, The React Framework for Production. [On- line]. Available: https://nextjs.org/docs

14. Hugging Face, Transformers: State-of-the-Art Natural Language Pro- cessing, 2024. [Online]. Available: https://huggingface.co/transformers

15. Ministry of Education, Govt. of India, National Education Policy 2020, Ofcial Report, 2020.

16. R. Patil et al., AI-Based Recruitment and Interview Automation: A Survey, International Journal of Computer Applications, vol. 185, no. 6, pp. 17, 2022.

17. R. Gupta and S. Verma, Resume Screening and Skill Extraction Using NLP Techniques, International Journal of Information Technology, vol. 14, pp. 145153, 2022.

18. N. Reimers and I. Gurevych, Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks, in Proc. of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2019, pp. 39823992.