MEDALLION ARCHITECTURE FOR

SCALABLE AI-DRIVEN INTERVIEW

PREPARATION AND REAL-TIME FEEDBACK
AROCKIA RAJ A 1, REPANA JYOTHI PRAKASH 1, MADDELA HARIPRIYA 1, NALAPAREDDY DINESH KUMAR REDDY 1,
KANDIKATLA BALAJI 1

Department of Computer Science & Engineering (Data Science)

[email protected]
[email protected]
[email protected]
[email protected]
[email protected]

Madanapalle Institute of Technology & Science, Madanapalle, Andhra Pradesh

----------------------------------------------------------------------***----------------------------------------------------------------------
ABSTRACT the system to generate a fully customized assessment suite which
would include 30 multiple-choice questions, two coding tasks,
The platform has been designed with an AI-driven approach and a proctored mock interview conducted in a virtual
toward personalizing interview readiness through automated environment. AI-driven tools undertake real-time transcription
assessment and insightful feedback. We start by assessing and summarization of the candidate's response during the mock
resumes and job descriptions, applying the concepts of text interview, gathering insights of extremely minute details to
summarization and keyword extraction to curate relevant communication and technical skills. Following the tests, the
content for each candidate. With the refined data in hand, the system evaluates the performance of each candidate, defines the
system dynamically generates a customized preparation toolkit: areas requiring skill gap fill-up, and provides online resources as
30 multiple-choice questions, two coding tasks, and a fully well as recommends targeted improvement. Built on a strong
virtual, AI-proctored mock interview. The alternative format technology stack, such as Python, Node.js, Flask, and AWS,
opens avenues for real-time transcription and summarization of within a Medallion architecture, the solution scaled and
the candidate responses during the interview. Results from these maximized efficiency in the making, giving it a benefitting range
exams are then analyzed to reveal knowledge gaps, where the of contenders and recruiters. The combination of automated skill
platform offers customized improvement suggestions along with assessment, AI proctored testing, and data-driven feedback
online resources which will guide the candidate on his learning brings forth a transformative approach for interview readiness
journey. Using Python, Node.js, Flask, and AWS, this will be preparation by equipping candidates with actionable insights
part of a Medallion architecture where it shall integrate for a that improve their readiness and success in their career pursuits.
scalable solution, both for the candidate and the recruiter, that
would better prepare candidates through interview round with
precision and efficiency via data-driven insights. 2. BACKGROUND

Key words: Many of the recent works focused on improving the interviewing
process and the recruitment by applying AI-based technologies
AI-driven interview preparation, Automated skill assessment, to fine-tune the preparation process. There were earlier efforts
Personalized candidate feedback, Proctored virtual mock that could be categorized as individual steps: automated resume
interviews, Medallion architecture for scalability screening, checking for basic skills, and keyword extraction
from job postings for matching candidate qualifications. These
1. INTRODUCTION efforts were able to give some preliminary insights, but they
were limited because they only covered one feature instead of a
In the modern job market, interview process complexities call complete flow. Where it most hindered the progressive
for candidates to be unusually well-equipped in a variety of development of earlier designs was its inability to include all-
skills. Even classical forms of preparation often fail to give around, individualized feedback and actionable
personalized suggestions and recommendations. In a bid to recommendations on actual domains where a candidate needed
bridge this gap, this study introduces a comprehensive AI-driven to improve; an improvement-for-all-one-size-fits-all that
platform for interview preparation tailored uniquely for every satisfied nobody's needs. Some additional developments were
candidate's qualification and the requirements of the target job. made in question auto-generation, coding tests, and even simple
The system first starts its preparation process through resume virtual mock interviews; however, the majority of the systems
and job descriptions collection and analysis with the application continued to rely on proctoring by individual or semi-automated
of advanced text summarization and keyword extraction that human agents, limiting scalability and compromising security.
distill relevant topics and competencies. Such refined data help Moreover, since machine learning and natural language
processing techniques were used in handling resumes and job 4. PROPOSED SYSTEM
descriptions, earlier systems often could not offer targeted and
actionable recommendations and sometimes did not provide 4.1 Cleaning and Structuring Data:
training materials or suggestions for closing the skill gaps that The process begins with collecting and fine-tuning two basic
had been identified. documents in the initial phase of preparing candidate
assessments: the resume of the candidate and the job description
3. LITERATURE SURVEY for that position. A resume will provide an overall outlook on
qualifications, experience, skills, and achievements of a
The field of AI-driven interview preparation has garnered candidate. Conversely, the job description outlines the specific
significant attention as organizations increasingly adopt competencies, responsibilities, and qualifications required in
advanced technologies to streamline hiring and training. that role. We first extract text from documents in various
Research into automated interview platforms shows that formats, such as PDF, DOCX, or HTML, and normalize this text
combining Natural Language Processing (NLP) with machine for further easy analysis. Normalization means converting the
learning enables efficient extraction and analysis of candidate text to all lower cases, eliminating special characters, and
information from resumes and job descriptions, facilitating removing stop words that add nothing to the value, thereby
personalized assessments. Studies by Kaur et al. (2020) and leaving a cleaner and uniform data set. We filter out extraneaous
Zhao et al. (2019) emphasize the effectiveness of NLP in resume pieces, such as the contact information on a resume or company
parsing and candidate-job matching, illustrating how text background in the job posting, to concentrate only on the
summarization and keyword extraction can reduce recruiter relevant issues that are at hand.
workload and improve hiring precision. Furthermore, the role of
automated question generation in training systems has been Finally, we extract critical features from both. This time, we try
explored extensively. For instance, Sun et al. (2021) to identify skills, experience levels, education, and
demonstrates that adaptive learning models, which adjust certifications. With NLP tools like Named Entity Recognition,
question difficulty based on candidate performance, can enhance we can specifically articulate skills and qualifications and
engagement and provide better skill evaluations. structure them into some kind of standard format, so it becomes
easy to compare the resume with the job description. That
The integration of AI-proctored systems is also an area of structured data is then turned into numerical vectors via TF-IDF
interest, with current research focusing on enhancing test or word embedding to make it machine learning friendly for
integrity and ensuring a controlled environment for remote further analysis, which enables the candidate profile to match
assessments. Studies, including those by Patel et al. (2021), semantically with the job requirements. This is therefore a more
reveal that AI-based proctoring solutions, such as facial refined data format that would improve accuracy in appraisal
recognition and activity tracking, play a crucial role in and provide a platform for more personalized career advice,
maintaining fairness and accountability, particularly in remote showing possible skills gaps and applicability to the job role.
testing scenarios. Similarly, research by Kim et al. (2022)
highlights the potential of virtual reality (VR) and augmented 4.2 Summarizing Text and Identifying Key Terms:
reality (AR) for immersive training and mock interviews,
helping candidates simulate real interview environments and
develop non-verbal skills.

Sentiment analysis and behavioral evaluation during virtual

interactions have emerged as promising tools for assessing
candidates’ soft skills, such as communication and emotional
intelligence. According to Sharma and Wong (2022), real-time
sentiment analysis during AI-driven mock interviews provides
valuable insights into candidate disposition, fostering a more
comprehensive understanding of candidate readiness beyond
technical knowledge.

Lastly, the application of big data analytics in identifying skill

trends is a growing area in talent acquisition research. Analytics
platforms are being used to predict emerging skills based on Fig 1. Keywords extractor
large datasets of candidate performance. For instance, the work We apply NLP techniques to distill information from both the
of Gonzalez and Lee (2023) suggests that data-driven insights resume of the candidate and the job description into key insights.
can significantly inform curriculum design for educational This is important as, on a one-to-one basis, resumes and job
institutions, allowing for more responsive workforce descriptions are big blocks of information, most of which are
development. By incorporating these advancements, the irrelevant to the job in question. From the text summarization,
proposed interview preparation system stands poised to leverage we can narrow it down to most crucial qualifications,
proven methodologies to enhance the efficiency, accessibility, experiences, and skills from the resume and the core
and precision of candidate training, positioning itself as a requirements from the job description. The techniques of NLP-
comprehensive tool for modern job preparation. based extractive summarization are used to capture key
sentences or phrases that reflect the major themes of each
document. Summary information forms a basis for comparison questions (MCQs), coding tasks, and a virtual practice
between the candidate profile and the expectations of the job. interview; all designed to assess different aspects of the
competencies of the candidate as suited to the requirements for
Keywords to be extracted: After summarization, the requirement the job. This is done based on NLP analysis; thus, every question
is to pull out keywords that represent the most significant skills, or task would directly relate to skills and qualifications stated in
qualifications, and qualities for the role. Techniques applied to the job description. The first component will consist of 30 MCQ
keyword extraction are Term Frequency-Inverse Document questions targeting basic knowledge of job-related skills that
Frequency (TF-IDF), Named Entity Recognition (NER), and the may include programming languages, tools, or industry best
more advanced model of BERT. Keywords from the resume practices.
include job-specific skills, certifications, industry terms, and
achievements manifesting the Candidate's expertise. Keywords They will probe the candidate's theoretical understanding and
in the job description highlight the kinds of skills, experiences, permit quick evaluations of their baseline skills. The second
and competencies that are prioritized in an employer. Aligning element includes two coding tasks that focus on real problem-
these keywords between the resume and the job description solving abilities, simulating real-life cases in the job. These tasks
ensures that the system identifies overlap and gaps in test coding, algorithmic thinking, and how good the candidate
qualifications and will be used in increasing the relevance and can write tight, clean code. Finally, the virtual practice interview
targeted evaluation of the candidate. The process ultimately is meant to simulate real-life interview environments. AI-
helps in giving recommendations on how well the background powered analysis reflects technical and soft skills such as
of the candidate aligns with the role and the points for problem-solving, analytical thinking, and verbal
improvement. communication, adaptability, teamwork, and interpersonal
skills. This integrated offering proves to be holistic for
4.3 Architecture Diagram evaluating the candidate's aptitude for the role, and hiring teams
will back the decision with data inputs.

4.5 AI-Supervised Evaluations:

AI supervises and ensures the integrity and fairness of the entire

evaluation process, such as MCQs, coding tasks, and virtual
practice interviews. The system ensures through AI-driven
proctoring that things run in a controlled and secure
environment, so there is no worry of cheating or other
unauthorized help. This kind of supervision is important when
test assessments are taken remotely, due to which direct human
supervision might not be possible.

Fig 3. Preparing tests with models

Fig. 2 Architecture Diagram
At the time of the MCQ and coding test, the AI proctor was
actively monitoring candidate activity in real-time using
4.4 Generating Comprehensive Evaluation Materials: techniques involving facial recognition, gaze tracking, and
screen monitoring to pick up unusual behavior. For example, it
All the structured and refined data from the resume and the job can detect events such as over-the-edge consultation of external
description is used in the building of a customized assessment sources or breaking eyes off from the screen. At the same time,
package. It prepares candidates for a realistic experience in an AI tracks screen opening by the candidates so that nothing
interview. The three main components are: multiple-choice unconnected opens during the test. However, the observation of
AI towards the virtual practice interview extends just beyond ensuring all details of their communications - their tone, choice
marking the candidate's response to pointing out general of words, and clarity of expressions - are captured and analyzed.
behavior throughout the interview. Through the translation of audio responses to text, the system
can make use of NLP techniques towards a better evaluation of
AI observes the verbal and non-verbal communications during the candidate's verbal skills through the assessment of relevance,
an interview, where it judges not just the content answers but the coherence, and conciseness in their answers. Transcription and
soft skills like confidence, articulation, and interest displayed in summarization in real-time
an interview. This AI tool, using natural language processing and
sentiment analysis, can evaluate the tone, clarity, and coherence Real-time transcription and summarization allow several
of the responses in order to ensure that the interview accurately functions. It enables the AI to feature key points from the
represents the candidate's capabilities. This AI-driven proctoring answers provided by the candidate for easier generation of an
will provide a standardized, unbiased evaluation process for objective opinion immediately. These summaries are very
every candidate while keeping the testing environment secure helpful to the hiring teams, as they present a brief version of
and fair. performance by the candidate, showing both strength and scope
for improvement. Also, transcription is effective for detailed
4.6 Audio Transcription for Virtual Mock Interview: analysis of soft skills, which include thought clarity, articulation,
Advanced speech-to-text technology is being integrated into and emotional tone aspects, if indispensable for tasks involving
virtual mock interviews which will allow the AI system to effective communication. It captures even subtle performances
transcribe and summarize in real time the responses given by the as it ensures proper manifestation of the interpersonal and
candidate through spoken words. This transcription process communicative skills of the candidate. This also simplifies the
allows creating a comprehensive record of all the answers the assessment process of the interview and provides a reliable and
candidate gives, so all aspects of communication - tone, unbiased record of the interaction that hiring teams can use for
vocabulary, clarity - can be captured and analyzed. The system informed decisions depending on the abilities of the candidate.
will automatically convert the audio response to text, where it 4.8 Customized Suggestions for Resources:
can apply NLP to develop a more in-depth evaluation of a
candidate's verbal ability in terms of relevant, coherent, and A system progresses from "Customized Recommendations for
concise responses. Resources" by providing specific, practical recommendations on
resources to bridge specific skill gaps, rather than suggesting
areas where the candidate needs improvement. It is only after
reviewing the candidate's performance that the system narrows
down to specific areas for improvement-whether there's a gap in
technical knowledge, inefficient coding, or poor communication
skills-and provides relevant learning resources. This focused
method makes the process of the feedback not only informative
but also transformational since the candidates are led to
understand how they can improve their abilities.

This is achieved through tapping into a vast digital learning

Fig 4. Voice to text convertor
material repository, comprising articles, online courses, video
Real-time transcription and summarization will serve several tutorials, and other interactive tasks. These resources are
purposes: first, because the AI is actually able to bring out key carefully curated from reputable educational platforms and
points during a candidate's responses, making it easier to providers for optimal quality and relevance. For instance, if the
produce an immediate, objective evaluation. These summaries candidate is particularly weak in some coding concept, the
are really useful for hiring teams, as they give the condensed system may recommend some online course or video series on
version of the candidate's performance that can focus on their that topic. If they need a reminder about interview
strengths as well as areas of improvement. Transcription also communications, it may suggest articles and practice exercises
allows for an in-depth analysis of soft skills involving clarity of focused on behavioral interviewing. Because the system is
thought, articulation, and emotional tone, which happen to be always working on specific weaknesses with the best available
principal requirements in most communication-intensive roles. resources, it empowers the candidates to do the same thing and
It captures, even the subtlest nuances, meaning that the ensures that feedback is actionable and facilitates continuous
candidate's interpersonal and communication abilities are improvement. This personalized support helps the candidate get
reflected accurately. This approach streamlines the appraisal of to their best, so when they finally go for those roles, they are
the interview itself but also forms a reliable record of the better prepared for them.
exchange that can be taken advantage of by the hiring team to
4.9 Advanced Tech Setup:
inform decisions based on candidate capability.
The "Advanced Tech Setup" ensures the system operates stably,
4.7 Evaluation Analysis and Feedback Creation:
is scalable, and delivers quality user experience. The
Advanced speech-to-text technology is integrated in the virtual architecture itself is the fusion of cutting-edge technologies that
mock interview process to enable the AI to transcribe and will both stabilize and flex the system. Python is at the core of
paraphrase the audio answers of the candidate. The transcription the back-end due to its processing capabilities coupled with
process provides a complete record of the answers of candidates, learning from data-With true power to process large volumes of
data and developing models for AI. These are so critically overtime. These platforms are multilingual and will serve a
required for candidate performance analysis, highlighting global population; even those who are not native English
valuable skills, generating customized feedbacks, which require speakers can access customized preparation in their native
some strong computational support. The system using Node.js language. Industry-specific content will ensure that the trainees
and Flask is relied upon for interactive users as well as content get targeted practice in whichever kind of field they are
delivery. Node.js will be another good asynchronous process interested in-be it finance, technology, or healthcare. Sentiment
management platform that allows for the quite efficient user analysis and soft skills assessment will hone effective
requests and content distribution on websites. Flask will communication, confidence, and emotional intelligence. Virtual
complement this with lightweight, flexible backend and augmented reality (VR/AR) will create realistic
development tools that will be good for API requests and routing environments for interviews where the candidates can practice
tasks. This integration of technologies enables a user-friendly non-verbal communication and even problem-solving activities
interface and the smooth execution of backend processes, in lifelike settings. That could be possible due to big data
providing an integrated experience for candidates and analytics, which outlines emerging skill gaps and provides
administrators. The use of Amazon Web Services (AWS) candidates with insights related to the most in-demand skills.
provides all the necessary cloud infrastructure in addressing Educational institutions and employers alike would receive
scalability and the system's computational needs. It scales the precious data related to where their training programs needed
system with no down time completely efficiently based on alteration. Because of its global scalability, the system might
changing loads and ensures there is high availability so that it is then adapt to various regions with their cultural nuances and
critical for real-time analysis and processing of large data sets. local job markets' specific requirements. With such
Furthermore, the advanced computational power availed from advancements, AI-based interview preparation systems are
the use of the cloud setup ensures that complex tasks such as destined to become the most basic tool for all potential job
natural language processing and real-time AI evaluations are seekers across the globe to empower themselves to become
handled more efficiently. successful in a growingly challenging and competitive job
market worldwide.

ACKNOWLEDGEMENT

We would like to extend our profound gratitude to Mr. Arokiaraj

A, whose mentorship, dedication, and insightful guidance have
been invaluable to this research journey. His expertise and
support have not only enriched the quality of our work but have
also inspired us to push the boundaries of our knowledge and
skill. His unwavering belief in our abilities encouraged us to
overcome challenges and approach each phase of this project
with confidence and determination. We are deeply thankful to
Fig 5. Medallion Architecture Dr. S. Kusuma, Head of the Department, for providing an
enriching and supportive academic environment. Her
This Medallion framework, also integrated into the design, commitment to fostering a culture of intellectual curiosity has
makes sure data flows both continuously and smoothened at all enabled us to conduct our research with a strong sense of
steps of the system. The framework standardizes the processing purpose and direction. Dr. S. Kusuma’s leadership has greatly
of data, meaning that data will move reliably from one module enhanced the learning atmosphere, allowing us to grow both
to another; beginning from initial data collection and summation academically and personally. Our sincere appreciation extends
to evaluation and generation of feedbacks. The Medallion to all the teachers in the department, whose wisdom and
framework sets a solid foundation for on-the-go updates and guidance have significantly contributed to our academic journey.
improvement, allowing developers to add new features or Their dedication to teaching and their passion for their subjects
expand existing ones without disrupting the core processes. This have equipped us with a strong foundation of knowledge, which
advanced setup positions the system for long-term adaptability has been instrumental in developing this research. The
and allows it to respond to emerging technologies and growing encouragement and support from the faculty have motivated us
user needs. to strive for excellence and to appreciate the values of hard work,
integrity, and collaboration. We would also like to recognize the
FUTURE SCOPE support and camaraderie of our friends and peers, who provided
The future of the AI-driven interview preparation systems is constant motivation, valuable discussions, and unwavering
immense and promising with innovations that will greatly encouragement throughout this process. Their companionship
enhance personalized learning, global accessibility, and and shared enthusiasm made the challenges more manageable
industry-specific customization. These systems measure and the successes more meaningful. Without their support, this
strengths, weaknesses, and learning preferences so that they can journey would not have been as fulfilling. Each of these
provide a tailored experience and continue adjusting the individuals has played a pivotal role in the successful completion
difficulty levels of the questions to reflect real-time of this research. Their encouragement, insights, and shared
performance. They will include adaptive learning models that passion for knowledge have been instrumental in shaping the
will challenge the candidates appropriately, along with outcomes of this work, and we are profoundly grateful for their
reinforcement learning for increasingly accurate feedback in contributions.
REFERENCES [11]

[1] https://ieeexplore.ieee.org/document/9753313

https://www.researchgate.net/publication/372441452_Appropri
ate_Job_Selection_Using_Machine_Learning_Techniques
[12]

https://ieeexplore.ieee.org/document/9753313
[2]

https://www.researchgate.net/publication/372663485_A_predic
tive_analytics_solution_matching_job_seekers'_talent_and_em [13]
ployers'_demands_based_on_machine_learning https://ieeexplore.ieee.org/document/9908817

[3] [14]
https://www.semanticscholar.org/paper/Job-Recommendation- https://www.ijraset.com/best-journal/elevating-performance-
System-based-on-Machine-Learning-Jain- through-ai-driven-mock-interviews
Kakkar/fe661340e332779f8c40dca713011f0fad938688

[15]
[4]
https://www.ijcrt.org/papers/IJCRT24A5067.pdf
https://www.researchgate.net/publication/378952763_Machine
_Learning-Powered_Cloud-Based_Text_Summarization
[16]

[5] https://ieeexplore.ieee.org/abstract/document/10208613

https://www.researchgate.net/publication/220974615_Automati
c_Text_Summarization_Using_a_Machine_Learning_Approac
[17]
h
https://abhivruddhi.mituniversity.ac.in/wp-
content/uploads/2023/08/Resume-Builder-.pdf
[6]

https://www.researchgate.net/publication/372261546_Online_E
[18]
xam_Proctoring_System_Based_on_Artificial_Intelligence
https://library.iated.org/view/BIRZANEANU2024REC

[7]
[19]
https://ieeexplore.ieee.org/document/10151160
https://ieeexplore.ieee.org/abstract/document/10063010

[8]
[20]
https://www.researchgate.net/publication/374004866_RETRA
CTED_ARTICLE_AI- https://www.irjmets.com/uploadedfiles/paper//issue_2_february
based_online_proctoring_a_review_of_the_state-of-the- _2024/49766/final/fin_irjmets1709095423.pdf
art_techniques_and_open_challenges

[21]
[9]
https://www.irjmets.com/uploadedfiles/paper//issue_10_octobe
https://ieeexplore.ieee.org/document/10151160 r_2024/62526/final/fin_irjmets1729401449.pdf

[10] [22]
https://ieeexplore.ieee.org/document/10170577 https://ieeexplore.ieee.org/document/10126233