Received: 30 January 2026; Revised: 20 March 2026; Accepted: 27 March 2026; Published Online: 30 March 2026.
J. Smart Sens. Comput., 2026, 2(1), 26204 | Volume 2 Issue 4 (March 2026) | DOI: https://doi.org/10.64189/ssc.26204
© The Author(s) 2026
This article is licensed under Creative Commons Attribution NonCommercial 4.0 International (CC-BY-NC 4.0)
An AI-Driven Training and Placement Platform with
Predictive Analytics and Conversational Assistance
Srikar Kulkarni,* Vaishnavi Kamthe, Kumar Saransh, Nemat Momin, Sonali Shirke and Mukul Jagtap
Department of Computer Engineering, Keystone School of Engineering, Pune, Maharashtra, 412308, India
*Email: srikarkulkarni49@gmail.com (S. Kulkarni)
Abstract
The increasing volume and diversity of student performance data have exposed significant limitations in traditional
training and placement systems, which primarily rely on static eligibility criteria, manual shortlisting processes, and
delayed communication mechanisms. These systems lack the ability to leverage predictive analytics, resulting in
limited personalized, data-driven insights to enhance student employability based on individual skills and
qualifications. Although several studies have applied machine learning techniques for job applicant ranking, most
existing solutions lack real-time integration, interpretability, and conversational support within placement systems.
To address these challenges, this study proposes an AI-driven education and placement platform that integrates
machine learning-based placement prediction with conversational assistance and intelligent job matching. The
system utilizes XGBoost for predictive modeling, Sentence-BERT embeddings for semantic skill representation, SHAP
for explainable insights, and Retrieval-Augmented Generation (RAG)-based chatbots to provide real-time guidance
and interview preparation support. The platform is implemented using FastAPI and deployed on cloud infrastructure,
with automated email notification systems enabling real-time user interaction. The proposed system was evaluated
using a dataset of 1,200 student records, incorporating academic, skill-based, and experiential attributes.
Experimental results demonstrate an accuracy of 89%, along with strong performance across multiple evaluation
metrics, including precision (0.88), recall (0.89), F1-score (0.88), and ROC-AUC (0.88). Additionally, the system
achieved low inference latency (<150 ms) and maintained stable performance under concurrent usage conditions.
Overall, the findings indicate that integrating predictive analytics, conversational intelligence, and scalable system
architecture significantly enhances placement decision-making, improves student guidance, and enables institutions
to adopt a more efficient and data-driven approach to managing placement processes.
Keywords: Placement prediction; Machine learning; Training and placement system; Conversational chatbot; Educational
data analytics.
1. Introduction
Digital placement portals in colleges and universities are now common but are largely limited to administrative
functions.
[1]
They mainly focus on record-keeping, eligibility filtering, and displaying job announcements for students
and employers. However, these systems lack analytical capabilities to predict student performance or hiring outcomes
based on available data.
[2]
Typically, this means that placement is related to static minimum academic standards and
manual judgment from an administrator, resulting in a limited ability to quickly respond to changing patterns of
recruitment and skill requirements.
[3,4]
The rapid growth of data-driven technology and machine learning has changed
the way we look at education. It provides educators with tools that allow them to intelligently analyze and interpret
vast amounts of data about students so that educators can better personalize learning experiences for each individual.
Recent studies confirm that many schools are implementing AI-based educational systems as a result of greater demand
for adaptive learning environments and data-informed management within institutions.
[5]
In addition, compared with
traditional statistical analyses, machine learning methods are effective at identifying patterns, predicting outcomes,
and improving educational results.
[6]
The challenge for educators is to more fully incorporate machine learning into
their existing placement systems for students in the workforce.
[7]
In an average institution, dozens to hundreds of
students undergo placement processing at once (if not thousands), often without any form of intelligence-based
solution or analytical ability, resulting in placement processing being performed manually or with static rules more
than Seventy percent of the time. As an example, eligible filtering, job matching, and student shortlisting all suffer
from inconsistencies and delays. In addition, the absence of any predictive modeling means that at-risk students cannot
be identified before they are too late and therefore not provided with adequate recommendations for personal
development training. With growing amounts of data being processed through placement systems, the deficiencies
experienced in present-day placement systems will only increase.
[8]
Therefore, scalable systems using artificial
intelligence to provide adaptive decision-making with real-time analysis will need to be developed to meet these needs.
Previous research has indicated that supervised machine learning placement predictions are moderately to very
accurate when they are based on historical student data.
[9,10]
However, many of these studies involve only offline
experiments using created datasets and have not been applied to actual operational placement systems. Because of
limited resources, many of these issues will be examined, such as real-time inference, data drift, the incorporation of
institutional workflows, and user interaction. Most predictive models tend to operate independently, without a means
to communicate, so that the analytical results can provide meaningful guidance to students and administrators.
[11,12]
A key limitation of current placement systems is that they do not have conversational and/or interactive support systems
to help with real-time queries and lower administrative overhead. Although web-based portals help to share
information in a centralized manner, they usually rely on asynchronous communications (e.g., email), thus causing
delays in communication and inconsistent distributions of placement-related information.
[13]
From the perspective of
software system evaluation, there is also a major research gap in that there is no analytical component that is integrated
into the overall assessment of placement-related services through both the use of interaction-based measures and the
assessment of deployment by users.
[14]
The present study examines how this research gaps can be resolved by evaluating the implementation and performance
of a fully integrated functional placement system using empirical approaches as opposed to design-centric approaches.
The major contribution of this work is the experimental evaluation of the performance of a machine learning-powered
predictive pipeline for predicting placement, which functions within a working integrated placement system and is
enhanced by interaction through conversation. Instead of proposing new conceptual architectures, the current research
evaluates real-world data to evaluate the predictive accuracy and system performance and the feasibility of the practical
use of integrated AI-based placement systems. This aim is to establish an evidence base for the performance of AI-
based integrated placement systems under real-world conditions, thereby creating a connection between predictive
modeling research and deployable educational technology solutions.
1.1 Overview of placement systems
The placement process serves as an important connection between academia and industry recruiters. Historically,
placement cells have used manual methods or simple database systems to compile students' data and periodically
update their eligibility lists and matching details.
[3]
While a number of universities have moved to web-based portals
for training records and recruitment calendars, most still act only as data repositories instead of providing intelligent
decision-support tools.
[4]
These systems also serve their limits via record storage, document uploads, and
communication notifications via email. The systems do not make use of the great amounts of student data collected
over countless academic years and thus have not applied predictive or analytic processes to the data.
[9]
The recent digital shift in higher education has prompted the adoption of learning management systems (LMS) and
online assessment tools that yield useful performance data.
[10]
Unfortunately, these data are often disconnected from
placement databases, limiting their use in employability analytics.
[11]
Studies, such as those in [6] and [7], have
recommended that centralized placement management systems need to transition from a static database to an intelligent
platform that generates insights that can be actioned by students, faculty members, and recruiters.
1.2 Need for AI-driven automation
The growing volume of student data and the evolving nature of corporate hiring trends have rendered manual
placement management systems inefficient and error prone.
[1]
With thousands of students applying to multiple
companies every semester, placement coordinators face difficulties in managing eligibility criteria, schedules for
interviews, and communications with recruiters in real time.
[2]
More traditional methods require continuous manual
updating and verification of placement processes, which reduces the accuracy of the data and the administrative
workload. As discussed in [11], most existing web-based platforms lack predictive insights or adaptive learning that
can study placement patterns and outcomes and constantly predict placement success.
Artificial intelligence (AI) and machine learning (ML) technologies have shown great promise in changing data-driven
decision-making. They allow organizations to look for insights in historical datasets and detect trends that conventional
analysis cannot identify.
[12]
For instance, supervised learning algorithms such as support vector machines (SVMs),
random forests, and logistic regression consistently demonstrate efficacy in predicting student performance and
employability.
[13]
These models are capable of evaluating factors such as academic results, technical skills, internship
experience, and certifications to predict a student's probability of being placed.
In addition, automation through AI-powered chatbots and recommendation engines will provide students and recruiters
with timely assistance, and roles are shifting to more rapid communication and increased engagement.
[15]
The
introduction of these systems to placement reduces manual dependency, increases accuracy, and enables scalable
placement across institutions. Automated systems, which use AI, evolve the placement system from simply a data-
management tool to become an intelligent ecosystem capable of making proactive decisions that align student skillsets
with industry needs.
1.3 Limitations of traditional approaches
Although traditional placement systems are commonly used in academic institutions worldwide, they present a
considerable number of constraints that impact efficiency, effectiveness and scalability. Most importantly, placement
systems operate as manual or semiautomated record-keeping systems focused on data storage and retrieval
[11]
without
incorporating intelligent aspects, such as predictive analytics or pattern recognition, needed for understanding trends
in student employability and outcomes. Thus, academic institutions are often unable to identify skills gaps or be
proactive in delivering training opportunities to students who are underperforming.
[12]
Second, traditional systems rely on administrative support to help verify eligibility, shortlist candidates, and update
records after a recruitment cycle.
[13]
In addition to being very time-consuming, all of these tasks run the risk of errors,
inconsistencies, and delays in communicating with companies. Furthermore, these systems do not integrate training,
performance analytics, or placement outcomes, leading to data fragmentation for management purposes.
[15]
A significant drawback is the limited dynamic adaptability. Traditional systems are not designed to accommodate the
current rapidly evolving recruitment and skill-based assessment processes introduced by the industry.
[16]
For example,
most platforms do not have real-time analytics or personalized feedback for students, depending on company
requirements. Furthermore, the majority of current placement management systems lack the ability to perform other
advanced authentication methods. Therefore, there is potential for issues with the recruitment and data security and
privacy of students in terms of information related to placement agencies or tools.
[11]
These challenges illustrate the need for an intelligent, secure, and adaptive AI-driven solution to optimize operations,
support prediction accuracy, and provide a systemic view of student readiness to transition into employment.
1.4 Motivation for the proposed system
The escalating challenges posed by campus recruitment and the increase in employer demand for employability-ready
graduates signifies the need for a smart, flexible, secure, and integrated placement framework.
[13]
Institutions are
seeking systems that not only store placement data securely and efficiently retrieved from different sources but can
also help draw insights and help guide students and employers. The concepts of artificial intelligence (AI) and machine
learning (ML) now enable institutions to automate key decision-making processes and move placements in a
predictive, interactive model as part of the placement management experience.
[15]
The rationale for the AI-Powered
Training and Placement Portal is based on the notion that all of the functions-student profiling, placement prediction,
skill analysis, and communication-are two features integrated within one platform.
Current models provide some partial resolution intended either for data storage or for analytics; however, very few
have included an end-to-end intelligent ecosystem.
[16]
The described system combines the power of ML algorithms
with assessments of prior academic data, analyzed training performance, and recruiter feedback to predict students’
likelihood of placement. On the basis of this prediction, the system may also recommend personalized training or
certification programs to maximize employability outcomes.
[17]
Coupled with the integration of a chatbot, interface
access is supported by immediate responses to student enquiries, notifications of eligible opportunities, and continuous
communication with the placement officer.
[18]
This integration is holistic and allows both the student and administrator
to experience cohesive, data-driven and interactive placement.
In addition to automation, the system addresses concerns of data credibility, transparency, and recruiter trust, which
are increasingly important in the wake of fake job postings and unverified recruiters.
[16]
Consolidating all placement
activities, as well as including data validation features, will help ensure greater trust and accountability among
stakeholders. Therefore, the paired use of predictive modeling, an interactive chatbot, and secure data management
provides the basis for this research, embodying a disruptive approach to leading modern education.
1.5 Research motivation and objective
The rationale for conducting this research stems from a growing dissociation between the availability of comprehensive
student data at the level of institutional operations and its limited use in actual placement decisions at the institutional
level. Institutions are turning to more digital platforms; however, they continue to use reactive and manual processes
that do not utilize predictive information to improve student success. Additionally, with the accelerated pace of change
in terms of industry requirements and increased competition for campus hiring, there is a dire need for systems that
not only produce predictive analytics to forecast whether candidates will meet placement expectations but also provide
personal data-driven, proactive interventions to assist students through their own individual career placement cycle.
This study is therefore driven by the need to develop and test a combined, integrated, operational, intelligent placement
system that uses predictive analytics, explainable decision-making, and real-time conversational support. The primary
objective of this study is to close the gap between stand-alone machine learning models and applicable placement
systems by demonstrating the viability, scalability, and success of an AI-based solution for real institutional settings.
2. Existing/similar work on training and placement systems
The integration of artificial intelligence (AI) and machine learning (ML) into placement management systems has
garnered attention as an emerging research space, with various models addressing predictive analysis and automation
in relation to student employability.
[19,20]
However, the scope of most candidate systems reviewed is narrow with respect
to the overall placement process, whether through academic analytics or a focus of either communication management
or registration management with placement systems. While this emergent area is being developed, a holistic framework
with prediction, training and automation that links the capabilities across functions within one intelligent portal is
conspicuously absent in placement management systems.
[21]
Therefore, it seems appropriate to discuss the need for a
truly complete and integrated adaptive solution to enable effective intelligent management of the complete placement
life cycle from data analysis to student engagement.
2.1 Machine learning-based placement prediction models
Applications of machine learning (ML) in predicting student performance and placement have grown from its vast
potential to analyze large datasets and find patterns that would otherwise be undiscovered.
[22]
Research has investigated
various ML algorithms, such as support vector machines (SVMs), decision trees, random forests, and logistic
regression, which all label students as “placed” or “not placed” on the basis of academic backgrounds and skills.
[23]
Original features were accounted for, such as cumulative grade point average (CGPA), attendance, project experience,
and internships, to predict placement and employability.
[24]
Srimathi et al.
[3]
used a decision tree-based method to produce classification rules to predict placement likelihood on
the basis of academic and technical characteristics. The system produced moderately accurate predictions but was
inflexible across institutions with different programs of study. The methods in [11] and [12] similarly used either a
random forest or a naïve Bayes model to improve accuracy and accommodate imbalanced datasets. While both sets of
methods improved prediction accuracy, they relied on static datasets, requiring manual retraining each time new data
were available.
Recently, researchers have begun to apply deep learning models to placement prediction, using neural networks to
learn from a variety of student features at the same time.
[25]
Researchers have also begun applying deep learning
techniques with advanced feature analysis to predict placement outcomes by utilizing high-dimensional data
representations and model interpretability techniques, which enhance both the accuracy and robustness of predictions.
In terms of feature importance, feature selection techniques based on explainable AI (XAI), and feature importance
selection techniques have demonstrated marked improvements in model performance while maintaining
interpretability across multiple data types.
[21]
While these systems have shown higher accuracy and robustness than
traditional models do, they require large labeled datasets and much more computing power.
[15]
Moreover, most of the
reviewed ML-based systems do not function in combination with the relevant institutional placement portal and fail to
connect predictive information with practical placement management or communication modules.
[16]
The results indicate that although models incorporating ML perform well in academic analytics, their disjointedness
and inflexibility of use hinder their application in the real world of institutional placement ecosystems. This creates a
need for an intelligent framework that incorporates ML prediction and combines it with automation, scalability and
real-time decision support.
[10]
2.2 Existing training and placement portals
Most educational institutions have built their own digital or web-based placement management portal to simplify
administrative work and support data handling.
[10]
Generally, these systems include modules to register students, post
jobs, post company updates, and manage applications through shortlisting workflows. Most such systems still use
static logic and do not include intelligent automation.
[11]
The typical process allows students to fill in personal
information, upload their resumes, and select to apply for drives, while recruiters can post job profiles and eligibility
information. Typically, these systems have linear and manual data flows between recruiters and students, and not much
analytical analytics is built in them.
[12]
Research identified in [13] and [15] highlights that current portals are primarily information management systems
rather than intelligent systems. Information management portals provide offers in terms of managing processes and
reducing paperwork by offering the ability to maintain structure and databases, but they do not take advantage of data
analytics to predict a student’s readiness or job fit. Communication from and between students, recruiters, and
placement officers who use the portals generally occurs via emails or announcements, leading to delays or inconsistent
information.
[16]
A further limitation noted in [12] and [15] is the lack of real-time data synchronization and data validation, which can
result in problems such as job postings being outdated, duplicate records, or students being listed as eligible when they
are not. Most systems also do not provide any safeguarding, creating the possibility of unauthorized access or
misinformation about students' education from unverified recruiters.
[11]
Last, most existing portals do not include
features that utilize skill-based recommendations or individualized training paths, creating a streamlined approach to
preparing students.
While these platforms are key developments in digitizing placement, they are still transactional, relying on operational
convenience rather than intelligence; the need for an AI-focused suite of analytics, automation, and interaction to
improve placement results while enabling institutional efficiency is increasing.
2.3 Comparative analysis of current solutions
A comparative analysis of existing systems suggests that while several approaches utilize machine learning (ML) or
web-based automation to improve the placement process, relatively few holistic solutions demonstrate integrated
analytics, communications, and scalability.
[11]
Systems that focus only on an ML-based prediction model can achieve
moderate accuracy but have no practical application in an institutional context.
[12]
Web-based placement portals that
efficiently monitor placement records do not pursue learning algorithms, which limits their usability and ability to
predict.
[13]
As noted in [7], prior placement solutions can be generally identified by three modalities: analytical models,
management portals, and hybrid systems. Analytical models typically rely on classification or regression algorithms
to predict placement but do not engage directly with a user-facing application. Management portals facilitate the
organization of records but remain static and do not offer decision-making capabilities on the basis of the data. Hybrid
systems attempt to combine prediction and management, but like existing systems, there is little to no real-time
interactive adjustment to a user’s input or the context of moving recruitment environments.
[16]
Moreover, research comparing several models reveals that the majority of placement prediction systems report
accuracies ranging from 75–85%, which is contingent upon the algorithm and dataset size.
[13]
While this is encouraging,
the models do not typically have feedback loops or adaptive retraining to improve predictions over time. Similarly,
regardless of whether feedback loops or adaptive retraining are used, existing placement portals cannot accommodate
multiple concurrent students and recruiters, which creates data inconsistency during peak times.
[12]
Furthermore, it is clear from their previous research work that the literature review revealed that none of the systems
assessed addressed two essential challenges (without repeating, i.e., distinctions that set it apart) faced by the
recruitment automated system in the literature review, which are recruiter validity, conversational AI, and personalized
learning recommendation within a single integrated platform.
[15]
The absence of consideration of such a comprehensive
approach presents an opportunity to develop a robust AI-powered system that involves the use of recruitment
prediction, security, and communication. AI-Powered Training and Placement Portal therefore aims to provide
appropriate means for joining the gaps with a model using ML-based predictive models or apps with chatbot
engagement and centralized data to provide improved transparency and workflow efficiency.
[4]
If these gaps are addressed, a more robust, scalable, and adaptable intelligent placement system can be developed to
effectively support data-driven decision-making and improve overall placement outcomes.
Through a thorough gap analysis of current systems, it has become clear that most of these solutions operate
independently and lack interoperability, falling short in their three key areas of prediction, management, and
communication. There is an absence of analytical models (i.e., focus on accuracy of classifications) that have been
developed for implementation in a true real-world environment (i.e., practical use). On the other hand, many of the
current web-based placement portals provide some useful administrative functionality; however, none of the
deployment architectures possess predictive/analytical capabilities or have been designed to learn and adapt over time.
Hybrid models have been developed as attempts to bridge this gap. Unfortunately, these hybrid systems often lack
real-time means of interaction between users and the system (i.e., real-time), provide little or no method of explaining
how the decisions were made (i.e., explainability) and lack a scalable deployment model for concurrent users (i.e.,
scalability). The proposed system combines predictive analytics (i.e., an analytical model that uses XGBoost), semantic
understanding of skills (i.e., Sentence-BERT embeddings), explainable AI (i.e., SHAP values for explaining how the
AI decision was made) and real-time conversational support via a retrieval-augmented generation chatbot all into one
comprehensive platform. Additionally, unlike existing solutions (both analytical and administrative), the proposed
architecture enables real-time job matching, personalized recommendations, and a scalable deployment model that
allows for concurrent usage. This integrated model addresses all of the main limitations of current systems found
within previously conducted research and therefore provides an all-inclusive, practicable AI-based placement
ecosystem.
3. Challenges in training and placement systems
Although the use of digital placement management tools continues to grow, the effectiveness and scalability of many
of these tools are limited by various challenges. While some digital placement systems provide data storage and record
management, few systems provide any analytical information related to student employability trends.
[3]
In addition,
many digital systems are not agile enough to accommodate constantly shifting industry requirements and skill
expectations.
[4]
As placement operations become even more complex (with hundreds of students and dozens of
recruiters-along with an abundance of real-time data), the shortcomings of placement management tools become even
more pronounced.
[9]
These issues may decrease efficiency and disallow institutions from using data-based decision-
making to positively affect placement.
[2]
3.1 Data Management and Scalability
With every year that passes, an increasing number of students, recruiters, and placement activities are taking place on
each campus. Institutions are seeing it becoming increasingly difficult to maintain data integrity and performance
efficiency and to complete these placement activities.
[3]
Some of these legacy systems rely on either local databases or
spreadsheets that require frequent manual updates, which leads to duplicate and disparate data entries.
[4]
There is no
effective centralized data handling and no automation for retrieving and/or analyzing placement records that can
support institutions during the recruiting season.
[9]
Scalability poses a major challenge when hundreds of students and recruiters try to use the portal at the same time.
Most existing portals crash, freeze, or experience data conflicts when users simultaneously attempt bulk entries, check
registrations, or update data in real time.
[10]
Many of these systems do not include appropriate backup and recovery
solutions to prevent data loss or corruption.
[11]
The inability of systems to successfully scale in terms of both storage
and the number of users accessing or updating the same type of information not only contributes to the diminished
reliability of the system but also lowers users' trust in web-based placement tools as a whole.
Automated data preprocessing, cloud integration, and distributed database management offer solutions to these
challenges through AI-based solutions.
[12]
These methods allow organizations to handle large amounts of data
seamlessly while managing consistency and availability. The incorporation of scalable architectures allows the system
to grow spatially in institutional growth without sacrificing performance to provide the basis for an intelligent and
robust placement management environment.
The system proposal is built upon the premise of appropriate support for several types of deployment (scalability and
generalization). The backend architecture (FastAPI-based service layer + cloud-enabled PostgreSQL database)
sufficiently handles concurrent requests from users during peak placement times. The horizontal scalability of the
system is the result of the design of the system architecture in a modular manner such that the prediction engine
component, the chatbot interface component, and the database layer can all be independently scaled. Because the
prediction of placements is carried out using XGBoost, the model has the expected ability to generalize well to
heterogeneous student datasets because it effectively captures the nonlinear relationships among the many features of
high dimensionality (i.e., numerical, categorical and embedded text). Furthermore, by using embedding-based
representations from Sentence-BERT, the possibility of generalizing to different skill description employment types
and job requirements is increased. Thus, the combination of architecture and modeling decisions results in a system
that provides a robust, scalable, and transferable infrastructure to many types of institutions regardless of the difference
in the distributions of their respective data and their placement dynamics.
3.2 Lack of predictive and analytical capabilities
An important downside of traditional placement management systems is that they do not have intelligent prediction or
analytical modules to read students’ background data to find meaning.
[10]
Most existing systems look at data primarily
for administrative data entry purposes-to determine what to register students for and if they are eligible
[11]
-not for
decision-making or predicting employability. This also means that institutions are basing readiness in the recruitment
process only on past placement outcomes or static academic criteria and risk being inaccurate or outdated.
[12]
Artificial intelligence (AI) and machine learning (ML) technology provide robust analytic potential that allows us to
uncover hidden patterns in student data, such as the relationships among academic performance, skill level, and
placement success.
[13]
It is incredibly rare for these analytics to be utilized in institutional placement systems. Placement
officers cannot take an active role in identifying at-risk students or personalized training recommendations without
some sort of predictive model. In addition, recruitment is limited in its ability to quickly shortlist a suitable candidate
who best meets their skill requirements without any analytics or screening.
[15]
Furthermore, institutions' inability to deploy visualization dashboards and trend analysis limit their ability to monitor
key longitudinal performance metrics such as skill improvement, training productivity, and placement conversion
rates.
[16]
This leads to an almost entirely reactive position rather than a proactive approach to student success. By
adopting AI-enabled analytics, institutions can improve their ability to evaluate placement probabilities and develop
an understanding of changing job market trends to ensure that training is consistent with job market expectations.
3.3 Communication and coordination challenges
Effective communication is at the heart of an effective training and placement process. However, the majority of
organizations still utilize manual or semiautomated communication methods such as emailing notifications to students,
issuing circulars, or simply posting new information on a notification board.
[11]
This is a reason for delays in
communication, missed opportunities, or simple disruptions in communication between students and recruiters and
placement coordinators.
[12]
For example, students are often notified late about company requirements, eligibility, or
their interview time. This results in confusion and a lack of interest when their time arrives for a recruitment drive with
the company.
[13]
Another problem arises when communication occurs through a variety of platforms. Students may be required to use
email to register, a messaging platform for communication updates, and a separate portal for uploading documents;
these platforms communicate that separate entities and organizations can easily increase complexity, leading to
miscommunication.
[15]
When there is no seamless matching interface, placement officers must manage well more than
two hundred student queries manually each day, which will require considerable time and effort. This reliance on
human interaction often creates bottlenecks for even larger recruitment events.
Additionally, in most conventional systems, real-time query resolution cannot be achieved. When students have
questions about a company’s eligibility, profile requirements, or training program format, they are dependent on
placement officers to answer their questions, and this adds ineffectiveness and delay.
[16]
An AI-based chatbot interface
helps eliminate those issues by providing 24/7 assistance, automating answers to repetitive common queries, and
delivering consistent information. This reduces the burden on the administration and increases the level of engagement
and accessibility across all stakeholders.
3.4 Concerns of security and authenticity
As the prevalence of online placement portals and electronic recruitment procedures increases, the issue of data
security and validation is becoming a priority for educational institutions.
[12]
There are now various online applications
used to store sensitive student data on databases that are frequently not encrypted or controlled.
[13]
Therefore, data
breaches, unauthorized access, or misuse are possible. Furthermore, no adequate authentication layers exist in many
long-standing placement systems for recruiters, providing room for fake or fraudulent job postings.
[15]
Over the past few years, there have been a number of cases in which unverified companies have taken advantage of
online portals to either collect student data or request payments under the guise of recruitment.
[16]
These recruitment
scams erode institutional credibility and result in students losing trust in the placement process. A significant challenge
is the lack of audit logs or traceability capabilities that can be used to monitor user activity and identify unusual activity.
[17]
To mitigate these risks, placement systems must incorporate best data protection practices such as encryption, role-
based access, and recruiter validation.
[18]
In addition, AI and ML can support the security of the system by flagging
unusual login activity or inappropriate behavior of recruiters in real time. In other words, a secure and transparent
framework is critical for ensuring credibility, trust, and fairness in the placement ecosystem.
Table 1: Comparative analysis of existing training and placement systems on the basis of prediction accuracy and automation
capabilities.
Approach [Ref.]
Focus Area
Prediction accuracy
Limitations
System [10]
Academic
performance
analysis
78%
Static dataset; manual updates
required
System [11]
ML-Based
placement
prediction
82%
No real-time data integration.
System [12]
Web-Based
placement portal
--
Limited automation; lacks
analytics
System [15]
Hybrid prediction
model
85%
No chatbot or adaptive feedback
mechanism
Proposed
system
AI + Chatbot
Integrated
placement portal
89%
--
3.5 Ethical considerations and bias mitigation
The use of machine learning algorithms in supporting employees' career placement decisions can pose some risks
regarding fairness, bias, and ethical data use. Several mitigation strategies have been incorporated into the system to
address these issues. The first strategy is that instead of providing a single or ranked prediction for placement decision-
making purposes, it instead provides placement probability estimates, resulting in final decision-making for individual
humans and reducing automated bias during the decision-making process. We utilize SHAP’s explanation approach to
provide a quantitative representation of how each feature (e.g., academic background, skill level, and experience)
contributes to the predicted output of the model, providing an avenue for transparency and interpretability in the
decision-making process of our model. As a tool for feature importance analysis and model interpretation, SHAP has
been effectively used across both supervised and unsupervised learning frameworks to determine which features are
influential but not negatively affect model accuracy.
[26]
Next, sensitive attributes such as gender, caste, or
socioeconomic status are not included as features for predictive purposes, further promoting fairness in predictions
made through this system. Finally, the domain alignment and job matching modules are built to provide guidance and
not eliminate students from being offered career placements, thereby creating a system that promotes equal opportunity
among students. These design choices result in a more ethical, transparent, and accountable placement framework
based on AI technology.
4. Proposed methodology
4.1 Dataset description
The dataset utilized within this research is from means of one of Kaggle’s public datasets. The title of dataset is
“Engineering Student Journey” and consists of over 1200 student records with significant detail about their academic
performance and technical skill level, in addition to their internship and placement results. The complete dataset
includes individuals from three different branches of engineering (Computer Science, Electronics, and Mechanical
Engineering), which provides for a broad range of representational characteristics of student profiles.
The dataset includes numerical and categorical attributes and contains semester-wise GPA, average GPA, total number
of backlogs, and total percentage of attendance, internship experience and participation in extracurricular activities
like technical clubs. The dataset also includes skill-based features, i.e., programming languages (Python, Java, and
C++) and domain knowledge (Machine Learning, Data Science). The dataset has two attributes that relate to
placements: placement status and offered salary (CTC in LPA). These two attributes can be categorized as the most
important attributes for employability.
This research will use the placement status (will find employment or not). Thus, this will be a supervised binary
classification model. In preparation for training the models, the dataset went through pre-processing by removing the
null values, normalizing numeric features and encoding categorical variables. Additionally, skill descriptions provided
in a text format were converted into dense embedding representations with the use of Sentence-BERT to capture
semantic relationships of those skills concerning job descriptions.
The dataset provides a realistic and structured representation of student academic progress and career outcomes,
making it suitable for evaluating machine-learning models in placement prediction tasks.
4.2 Predictive analytics and student profiling
In the new system’s overall workflow, there are four parts of the project: the data collection phase, the data processing
phase, the prediction phase, and the user interface/interaction phase. The first step is to collect information about
students from institutional databases, which includes academic history, technical skills, internships (if applicable), and
extracurricular activities. The next step is to perform data preprocessing, which consists of normalizing numerical
feature columns, encoding categorical feature columns, and transforming textual feature columns (such as skill and
experience descriptions) into their dense vector representations using Sentence BERT embeddings. The final processed
feature set is then passed to the machine learning prediction module, where an XGBoost classifier produces the
probability of placement for each student. The predictions produced by the prediction module are then utilized by the
job-matching engine to identify relevant opportunities for each student on the basis of semantic similarity to the
opportunities posted within the system's database as well as domain alignment with career paths of the students whose
data were submitted to the system. Moreover, the chatbot interface provides users with real-time assistance, the system
continuously updates current placement records, and the system sends out automated notifications when placements
are made for students. This creates an integrated process for prediction (through the prediction module),
recommendation (through the job-matching engine), and communication (via the chatbot or automated notifications)
as part of one unified platform.
4.3 Mathematical model
The model for predicting placements is described as a supervised binary classification problem, where the input feature
vector for each student is defined as follows:
X = {x₁, x₂, x₃…....xₙ}[ʙ] (1)
The input features all relate to academic achievements, individual abilities, experience acquired through internships
and so forth. The function that XGBoost will produce for the prediction can be expressed as an ensemble of decision
trees:
f(X) = ΣTₖ(X)[𝒇(𝑋) = Σ_{𝑘=1}^{𝑁} 𝑇_{𝑘}(𝑋)] (2)
where Tₖ is the kᵗʰ decision tree, and the output of the ensemble of trees is passed through a sigmoid function to produce
a probability score that indicates whether the user will be placed.
P(placement) = 1/(1 + e^(-f(X))) (3)
To match students to jobs, semantic similarity can be obtained by using the cosine similarity between the skill vectors
of students and the job descriptions.
Similarity = cos(E_student, E_job) (4)
where E_student and E_job are embedding vectors that have been generated by Sentence-BERT. To calculate the final
matching score, we use the following equation:
Match Score = 0.7 × Similarity + 0.3 × Domain Alignment (5)
These formulas serve to present the predictive and recommendation functions of the proposed system in a mathematical
manner.
5. Possible approaches to overcome these challenges
The problems that are evident with conventional training and placement systems suggest that there is increasing
demand for an intelligent, scalable and secure option. In response to these issues, researchers have suggested the
adoption of artificial intelligence (AI), machine learning (ML) and automation frameworks as part of the institutional
placement process.
These types of technologies are utilized to provide insights into data analytics, predictive modeling,
and adaptable learning mechanisms that assist institutions in discerning trends and making informed decisions.
AI-assisted models are capable of analyzing large quantities of past students' data, discovering meaningful patterns,
and predicting employability outcomes with a high degree of accuracy.
Moreover, automation via chatbots, cloud-
based storage, and predictive analytics alleviates the need for human intervention while maintaining transparency and
quickness.
A more comprehensive approach to manage the workplace placement ecosystem that serves all
stakeholders-students, recruiters, and administrators-would be to combine machine learning algorithms with
enhancements in communication and security.
5.1 Predictive analytics and student profiling
A very promising methodology for overcoming the restrictions of traditional placement systems is the implementation
of predictive analytics to assess and predict student employability. Predictive analytics is the application of a machine
learning (ML) process to historical data to recognize patterns that shape placement outcomes.
[4]
Attributes such as
academic performance, skills proficiency, certifications, extracurricular involvement and internships are scrutinized to
determine the likelihood of a student being placed.
[9]
Algorithms (e.g., support vector machines (SVMs), decision trees, and random forests) have been successfully utilized
for the placement prediction of students on the basis of classifications.
[10]
In these models, correlations between skillsets
and employability success are drawn from previously placed and unplaced student records. Following training, the
placement probability of students is determined, followed by providing recommendations on specific ways of
improvement.
[11]
Unsupervised learning approaches (e.g., clustering) can also be used for grouping students who share
similar characteristics. Institutions are then able to adapt training programs according to a specific cluster of learners.
[12]
Student profiling using AI can go beyond statistical data analysis. This technology helps map a dynamic representation
of each student's journey and skill development experience measured over time. The system integrates a student's
profile or data with new indicators of performance, such as project evidence, certificate evidence, or mock interviews.
The platform can then provide tailored reminders on how to upskill or specialize in a particular domain.
[13]
This
adaptive, data-driven approach can help placement departments establish skill gaps sooner and improve their readiness
for recruitment, ensuring that every student has the support they need at the right time.
5.2 Chatbot integration for communication and guidance
Introducing a chatbot interface to the training and placement system may enhance communication efficiency,
accessibility, and user interaction. Common communication channels (such as emails and notices) may cause delays
and misunderstandings during the placement cycle.
[2]
The AI chat box provides students, recruiters, and placement
officers with an interactive, centralized method of exchanging information in real time.
[11]
Chatbots that incorporate natural language processing (NLP) can understand the user question, generate timely
responses, and access educational content to help students register a business, assess eligibility, and schedule
interviews.
[12]
Natural language processing (NLP) enables chatbots to understand what users are asking, provide
responses on the basis of their conversation history, and engage with users in real time across several different topics.
Research shows that conversational artificial intelligence (AI) platforms improve user engagement and responsiveness
in education and service-oriented businesses through the use of interactive communication with the unique identity of
the user.
[27]
Conversational interfaces are dynamic rather than static web pages. Conversational models create a two-
way conversation that produces personal, contextual responses for the user. The system can provide proactive updates
to students about new job openings, deadlines, and shortlist notifications instead of relying on announcements and
even manual outreach via email.
[13]
In addition to reducing administrative burden and offering 24/7 support,
[15]
the chatbot can operate as an effective
virtual placement advisor in managing the repetitive and frequently asked questions that occupy invaluable time for
the administrative staff. One study
[16]
revealed that when a chat assistant was properly embedded into the teaching
system, it was able to increase the user satisfaction, retention of information, and transparency of the institution. The
chatbot provides a central component for a more user-driven, interactive, and efficient placement experience for the
user and graduates, as it covers communication, guidance, and support, all in one interaction.
5.3 Security and data validation mechanisms
Ensuring the safety and authenticity of data is the utmost priority for the acquisition of trust in any digital system for
placement. Placement portals, which contain sensitive data, including student records, academic grades, and
credentials for companies, must have sufficient mechanisms in place to safeguard against the disclosure of data access
or data misuse.
[15]
Traditional systems often depend on measures of security that utilize basic authentication, which are
straightforward enough that they leave data exposed to breaches or manipulation by cybersecurity threats of any sort.
[13]
Security models supported by artificial intelligence provide smart and adaptive protection. The models support
proactive monitoring of user behaviors, identify abnormal behaviors and discover possible threats (for example,
password theft and violations of recruitment behaviors).
[15]
Moreover, data encryption and role-based access control
restrict access to sensitive data to only those authorized to access it while protecting confidentiality and system
integrity.
AI-supported verification procedures can additionally improve the recruiter verification method, where an algorithm
is used to evaluate the legitimacy of recruitment using the registration information of the recruiter, previous sources of
recruitment, and the similarity of job content posted.
[16]
This helps prevent fake companies or job postings from
capturing student information. Moreover, regular audits of the database will be conducted along with automated
backups to protect against data corruption or loss from accidents.
[17]
With these intelligent security components built in, an institution can create a trusted digital space that protects user
data and increases the confidence of the institution's students to those who recruit them. The combination of AI-based
monitoring and validation allows for greater transparency and reliability in these systems overall.
5.4 Integration of cloud-based infrastructure
The adoption of cloud-based infrastructure has become an effective and scalable method for managing the large and
dynamic volumes of data associated with training and placement systems.
[13]
On-premise servers and local storage
architectures are often bound by performance constraints, especially with mass data access, registration, or placement
drives. Cloud computing offers the capacity for on-demand scalability, allowing institutions to provision resources on
the basis of system load and user demand.
[15]
The ability of cloud computing to scale on demand is what makes it
possible for those systems to have the ability to dynamically allocate computing resources on the basis of the demands
of the users and the current system load. Cloud-based architecture has also been found in recent publications to have
a significant effect on the reliability, fault tolerance and real-time data processing abilities of the overall system, which
makes cloud computing a better fit for large-scale, data-intensive applications such as Intelligently Placing Systems.
With the transfer of placement management systems to a cloud service, institutions are able to provide high availability,
remote accessibility, and quicker data processing.
[16]
The use of cloud services connecting stakeholders-students,
recruiters, and administrators- allows the system to update in real time to ensure everyone is informed of any updates,
notifications, or changes in a dataset. Cloud platforms also come with built-in reliability and redundancy features such
as automated backup, data recovery and load balancing, which minimize downtime in line with the recruitment
process.
[17]
Cloud infrastructure and AI capabilities of analysis incorporate both distributed computing and storage, which are
often necessary for managing the scale of data needed for predictive modeling and student profiling.
[18]
Cloud
capabilities and architectures are experientially flexible, allowing institutions the options for a hybrid deployment that
couples local control with scalability across the world. In closing, the cloud environment enhances the current learning
environment to a new realm of efficiency, security and scalability as technology evolves and organizations respond to
changes that impact educational institutions.
Fig. 1: System architecture of the proposed AI-powered training and placement portal showing the interaction between the
user, admin, chatbot, and data management module.
The architecture of the system includes several parts that connect to each other to allow users (students/recruiters) to
perform real-time placement events. Users interact with the system through a web interface, and the front end
communicates with a back end designed using FastAPI with a centralized processing layer (the back end). The back
end will utilize various modules, including a module that uses XGBoost, a module that matches skill levels using
Sentence-BERT, and a chatbot that uses RAG to converse with users. These modules access a centralized PostgreSQL
database. In addition, the entire system is hosted on the cloud infrastructure to enable scalability, reliability, and
efficient support for many concurrent users.
Fig. 2: Architecture diagram
The architecture of the system uses a layered approach to provide scalability and modularity. The client layer contains a web
interface for users to interact with the system. All requests to the system are routed through an API Gateway (using FastAPI)
that will manage communications between the Frontend and Backend services of the application. The back end of the system
has multiple functional services. The prediction service uses the XGBoost model to predict the likelihood of placement for
each student. The embedding service uses sentence-BERT as the underlying technology for semantic skill matching. Finally,
the chatbot service uses a retrieval-augmented generation (RAG) pipeline to allow users to interact with the service in real
time. Each of these services accesses the same central PostgreSQL database that contains student profiles, job information,
and placement history. In addition to the previously described services, a separate email notification service is available to
support automated communication with users of the system. Finally, the entire system is hosted on a cloud infrastructure
platform to provide high availability and scalable services while supporting the efficient handling of multiple simultaneous
users.
Fig. 3: Activity diagram
The activity diagram shows every step of a student’s journey throughout the training and placement system, including
going from their first visit through submitting applications and being notified about their results. The journey starts
with being authenticated as a user so that existing users can go directly to the dashboard and new students must go
through the account creation process prior to proceeding. When students are authenticated, they will be able to update
their profiles and the skills that they possess, which will then be used by the system to generate job matches for them
on the basis of those profiles. The system then determines if any suitable jobs are available for those students. If so,
students will have access to view the available jobs and will have the opportunity to submit an application for those
jobs. Once a student submits an application, both the job posting and the application will be reviewed by the placement
authority before an applicant is either notified of their eligibility for an interview or advised that they will not be
considered for further consideration on the basis of their qualifications. Upon completing an interview or applying for
a job, students receive a confirmation email with their application or an order when they apply for work. At this point,
the cycle of placement has been closed. This activity flow demonstrates that there is a major emphasis placed on
controlling user access on the basis of the role of the user, how to make decisions on the basis of conditions, and how
to communicate with users in an automated manner throughout the placement process.
6. Results and discussion
A comprehensive set of model-level and system-level metrics was employed to evaluate the performance of the
proposed AI-based placement system, with comparisons made against existing baseline methods. Traditional
placement systems and earlier machine learning models typically rely on static datasets and limited evaluation metrics,
with reported accuracies generally ranging between 75% and 85%.
In contrast, the proposed model was evaluated against baseline algorithms such as Logistic Regression and Decision
Trees, consistently demonstrating superior performance across all evaluation metrics. The proposed XGBoost model
achieved an accuracy of 89%, along with strong performance in terms of precision (0.88), recall (0.89), F1-score (0.88),
and ROC-AUC (0.88). These results indicate the model’s ability to generate consistent and reliable predictions for
student placement outcomes.From a system-level perspective, the proposed architecture achieved an inference latency
of less than 150 ms under concurrent user conditions, highlighting its suitability for real-time deployment-an aspect
often lacking in previously developed systems. Furthermore, the incorporation of Sentence-BERT embeddings
enhanced semantic alignment between student skill sets and job requirements, thereby improving job-matching
accuracy. The integration of SHAP-based explainability further strengthened model transparency by providing
interpretable insights into feature contributions.
Additionally, the combination of an Interactive Experience Management System (IXMS) and a Retrieval-Augmented
Generation (RAG)-based chatbot significantly enhanced system interactivity by delivering real-time, context-aware
guidance and company-specific insights. Recent advancements in generative AI indicate that RAG-based systems
improve response accuracy, interpretability, and contextual relevance through external knowledge retrieval and natural
language generation capabilities.
[28]
Collectively, these results demonstrate that the proposed system not only improves
predictive accuracy but also enhances usability, scalability, and practical applicability, thereby establishing it as a
comprehensive and intelligent placement solution.
Table 2: Confusion matrix of proposed XGBoost model.
Predicted: Placed
Predicted: Not Placed
Actual: Placed
520
60
Actual: Not Placed
70
550
Table 3: Comparative analysis of existing training and placement systems on the basis of prediction accuracy and automation
capabilities.
Accuracy
Precision
Recall
F1 - Score
ROC - AUC
78%
0.76
0.75
0.75
0.80
82%
0.80
0.79
0.79
0.83
89%
0.88
0.89
0.88
0.88
The ROC curve was employed to evaluate the classification performance of the proposed model by visualizing its
behavior across different threshold levels. The curve demonstrates a favorable balance between the true positive rate
(sensitivity) and the false positive rate, indicating the model’s strong ability to distinguish between placed and non-
placed students. The Area Under the Curve (AUC) value of 0.88 reflects a high level of classification performance.
Overall, the ROC curve highlights the robust discriminative capability of the proposed integrated system.
Fig. 4: ROC curve of proposed system.
7. Enhancements in training and placement through ai integration
Several performance measures were analyzed at the individual model and overall system level to conduct an all-
encompassing evaluation of the proposed system. A prediction model created using an XGBoost classifier will have
been tested for standard classification metrics of prediction accuracy (i.e., classification accuracy, precision, recall,
and F1 score), along with the ROC curve or AUC, to ensure that performance is balanced across each of the
classifications. The prediction model's predictive accuracy can also be examined via SHAP, which can enhance our
understanding of how each variable contributes to prediction or feature performance (e.g., academic performance, skill
embedding, and industrial experience [Internship]). The system-level performance metrics also assess the model's
ability to generate predictions in a timely and efficient manner. For example, the average time to produce a prediction
remains less than 150 ms when it is predicted simultaneously (real-time). Together, these performance metrics provide
a comprehensive evaluation of the prediction performance as well as the deployment and usability of the system.
Additionally, this multidimensional evaluation approach provides an accurate picture of both prediction performance
and deployment/capacity performance, which is something that has not been accomplished with existing research
because of the use of singular evaluation approaches.
Artificial intelligence (AI) and machine learning (ML) transform training and placement systems through solutions to
manage the analytics of employability, recruiting, and student readiness. Typical systems are limited to static data
management, whereas AI-enabled systems learn continuously to better inform decision-making through student
metrics and by leveraging recruiter feedback and placement results.
[3]
By automating prediction, communication, and
analysis of results, AI enhances the speed and quality of placement operations.
[4]
Additionally, when converged with intelligent chatbots, predictive algorithms can provide real-time interaction as well
as customized recommendations to aid students in measuring their strengths and weaknesses and aligning their profiles
to relevant job possibilities.
[9]
AI-enabled insights help administrators by indicating performance trends, noting skill
development gaps, and directing comprehensive training to remedy this deficiency.
[10]
Ultimately, this creates a more
dynamic and data-driven ecosystem that grows and improves with each placement cycle, ultimately leading to further
transparency and continuous improvement among all stakeholders.
[11]
7.1 Improved placement prediction and decision-making
The use of machine learning algorithms improves the accuracy and effectiveness of placement prediction models. The
system is able to process multiple pieces of information on the basis of the skills needed (e.g., academic performance,
technical skills, relevant certifications, and interview performance) to determine the most important factors leading to
successful placements.
AI models learn dynamically with new points of reference and make predictions instead of
learning in a static manner with a list of criteria as a general eligibility list, as conventional models do.
[29]
This adaptive prediction system allows placement officers to make better decisions on training needs and company-
specific shortlisting. The model analyses historical data to project which students are best suited to meet specific
recruiter needs. It helps recruiters by automating the search for potential candidates and eliminating the time to
manually evaluate larger datasets.
The AI module of the proposed system uses supervised learning algorithms, including decision trees and random forest,
which work well for predictions that require classification. As more placement data are introduced over time, the
accuracy of the models improves, making them self-improving decision-support tools. By providing these predictive
abilities in the placement experience, organizations can pivot from once reactive placement tracking to a more
proactive talent readiness and opportunity matching approach.
7.2 Intelligent chatbot integration and real-time interaction
The implementation of AI chatbots is a significant movement toward increased usage of automation and access in the
training and placement process. More traditional means of communication such as emails can similarly create delays,
lacunae, or a lack of updates for student recruitment campaigns. Chatbots do streamline and help to bypass these
constraints by providing students, placement officers, and recruiters with real-time on-demand access to an exchange
of information.
[30]
Chatbots can also help respond to some of the most common questions in real time, around the
eligibility to apply to a company, registration deadlines, or time of interviews, all while decreasing some of the burden
placed on staff for documentation and administration.
The bot is capable of interpreting user questions and generating context-based responses using NLP. That is, it can be
characterized as an agent rather than a helpdesk service. The bot can also push notifications to students, such as
announcements about upcoming interviews, outstanding skills sessions, or company updates of relevance to its users.
This active engagement keeps students updated and engaged with relevant information throughout the recruitment
process.
[31]
The chatbot system serves a dual function by facilitating user communication while simultaneously functioning as an
intelligent system for feedback collection. The system allows users to input data after their interview, which creates
the possibility of analyzing student data and generating results that support the system in performing analytical
evaluations. The system improves and enhances performance through the collection of ongoing feedback, which
signifies training programs and improves the AI prediction model. The chatbot system even provides ongoing user
assistance, which leads to increased user satisfaction, and builds a student-centered placement system that operates
with full transparency.
[32]
7.3 Adaptive learning and continuous system improvement
A considerable benefit of deploying artificial intelligence (AI) in placement systems is its ability to continuously learn
and adapt. Traditional systems are generated once implemented and remain the same throughout the duration of the
system's use; on the other hand, the AI platform builds and evolves as new data and users engage with the system.
Students participate and engage in recruitment cycles, and the system captures outcomes, feedback, and performance
data during the cycle that can be used postrecruitment to improve and readjust the students' predictive models and
communication strategies.
[33]
As adaptive learning continues, the accuracy of recommendations and placement prediction that the system has learned
from its previous experience will become increasingly precise. For example, once recommendations identify patterns
that lead to placement success-such as a specific skill set, project experience or a certification-it will prioritize those
same items as it analyzes future options. In contrast, if the model identifies patterns of rejection, it can provide
administrators and students with the opportunity to address those weaknesses through specific training sessions.
In addition, the addition of chatbots provides opportunities for continual improvement through the collection of
qualitative data from users. Information such as student satisfaction, provisioning student reactions, or recruiter
feedback could likewise allow for retraining the AI models, improving the user experience and prediction accuracy.
Updates of AI models based on feedback make the placement portal a self-improving system that can support
institution-wide strategies for navigating an evolving industry environment.
[34]
The proposed system goes beyond automating the placement decision process by leveraging adaptive learning to
develop an intelligent and evolving platform that continuously enhances decision-making, communication, and student
success rates.
8. Conclusion
This study presented an AI-driven training and placement platform that integrates machine learning–based prediction,
semantic skill analysis, explainable AI, and conversational intelligence within a unified system. The primary objective
was to address the limitations of traditional placement systems by introducing a scalable, data-driven framework
capable of improving decision-making and student outcomes. The experimental evaluation validated the effectiveness
of the proposed approach. The placement prediction model, implemented using XGBoost, achieved an accuracy of
89%, along with strong performance across precision (0.88), recall (0.89), F1-score (0.88), and ROC-AUC (0.88). In
addition, system-level evaluation demonstrated real-time responsiveness, with inference latency maintained below 150
ms, confirming the feasibility of deployment in practical institutional environments. The integration of Sentence-
BERT embeddings improved semantic alignment between student skill sets and job requirements, thereby enhancing
job-matching accuracy. Furthermore, SHAP-based explainability increased transparency by providing interpretable
insights into model predictions. The inclusion of a Retrieval-Augmented Generation (RAG)-enabled chatbot enabled
real-time, context-aware interaction, supporting students with company-specific preparation and personalized
guidance. Overall, the results demonstrate that the proposed system effectively bridges the gap between standalone
machine learning models and real-world deployable solutions. By combining predictive analytics with intelligent
interaction mechanisms, the system offers a comprehensive, scalable, and transparent framework for next-generation,
data-driven placement ecosystems.
CRediT Author Contribution Statement
Srikar Kulkarni: Conceptualization, Methodology, Writing-Original draft, Writing-Review & editing. Vaishnavi
Kamthe: Methodology, Software. Kumar Saransh: Data curation, Formal analysis. Nemat Momin: Investigation,
Validation. Sonali Shirke: Supervision, Writing - Review & editing. Mukul Jagtap: Project administration,
Supervision.
Funding Declaration
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit
sectors.
Data Availability Statement
The dataset used in this study is publicly available on Kaggle (the “Engineering Student Journey” dataset). It was
utilized for research purposes and preprocessed for model training. The processed dataset and related materials are
available from the corresponding author upon reasonable request.
Conflict of Interest
There are no conflicts of interest.
Artificial Intelligence (AI) Use Disclosure
The authors confirm that no artificial intelligence (AI)-assisted technologies were used in the writing of the manuscript,
and no images were generated or manipulated using AI. AI-based tools were used solely for language editing to
improve grammar, clarity, and readability, in accordance with journal policy. The authors take full responsibility for
the accuracy, originality, and integrity of the work.
Supporting Information
Not applicable.
References
[1] D. Magdalenić, L. Luić, Assessing the impact of digital tools on the recruitment process using the design thinking
methodology, Administrative Sciences, 2025, 15, 139, doi: 10.3390/admsci15040139.
[2] Ç. D. Ertuğrul, S. Bitirim, Job recommender systems: a systematic literature review, applications, open issues, and
challenges, Journal of Big Data, 2025, 12, 140, doi: 10.1186/s40537-025-01173-y.
[3] A. Billig, J. Gottschick, K. Sandkuhl, Evolution of web computing systems: experiences from web-portal projects,
Proceedings of the 2005 31st EUROMICRO Conference on Software Engineering and Advanced Applications
(EUROMICRO-SEAA'05)., 2005, 223–230,doi: https://doi.ieeecomputersociety.org/10.1109/EURMIC.2005.27.
[4] N. Srivastava, M. Tripathi, V. Rai, The development of a job portal to facilitate in campus placement, 5th
International Conference on Advances in Computing, Communication Control and Networking (ICAC3N), IEEE,
2023, 1549-1553.
[5] E. Elbasi, M. Nadeem, Y. I. Alzoubi, A. E. Topcu, G. Varghese, Machine Learning in Education: Innovations,
Impacts, and Ethical Considerations, IEEE Access, 2025, 13, 128741-128770, doi:
10.1109/ACCESS.2025.3590134.
[6] Z. Ersozlu, S. Taheri, I. Koch, A review of machine learning methods used for educational data, Education and
Information Technologies, 2024, 29, 22125–22145, doi: 10.1007/s10639-024-12704-0.
[7] Y. Aljemely, Challenges and best practices in training teachers to utilize artificial intelligence: a systematic review.,
Frontiers in Education, 2024, 9, 1470853, doi: 10.3389/feduc.2024.1470853.
[8] R. Qureshi, P. S. Lokhande, A comprehensive review of machine learning techniques used for designing an
academic result predictor and identifying the multi-dimensional factors affecting student's academic results, 2024
2nd DMIHER International Conference on Artificial Intelligence in Healthcare, Education and Industry
(IDICAIEI), Wardha, India, 2024, 1-6, doi: 10.1109/IDICAIEI61867.2024.10842901.
[9] R. Srimathi, J. Naskath, B. A. Mathavan, T. Archana Pown, M. S. Rabiya, "Institution management system: student
module, 2022 Fourth International Conference on Cognitive Computing and Information Processing (CCIP),
Bengaluru, India, 2022, 1-8, doi: 10.1109/CCIP57447.2022.10058674.
[10] S. S. Sakthy, G. A. Macriga, J. A. Jasmine, V. V. Babu and N. M. Sayhanuddin, integrated web application for
skill development and job application, 2021 4th International Conference on Computing and Communications
Technologies (ICCCT), Chennai, India, 2021, 106-110, doi: 10.1109/ICCCT53315.2021.9711801.
[11] M. Babu, K. Sandhiya, V. Preetha, S. Sankara Eshwari, M. R. Chitra, Design of alumni portal with data security,
2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC),
Coimbatore, India, 2021, 1-7, doi: 10.1109/ICESC51422.2021.9532986.
[12] R. K. Kousik, G. Nagappan, Computer human interface for placement management system, 2024 IEEE
International Conference on Computing, Power and Communication Technologies (IC2PCT), Greater Noida,
India, 2024, 1245-1248, doi: 10.1109/IC2PCT60090.2024.10486671.
[13] V. Pavani, N. M. Pujitha, P. V. Vaishnavi, K. Neha, D. S. Sahithi, Feature extraction based online job portal,"
2022 International Conference on Electronics and Renewable Systems (ICEARS), Tuticorin, India, 2022, 1676-
1683, doi: 10.1109/ICEARS53579.2022.9752295.
[14] M. R. N. King, S. J. Rothberg, R. J. Dawson, F. Batmaz, Bridging the edtech evidence gap: A realist evaluation
framework refined for complex technology initiatives, Journal of Systems and Information Technology, 2026, 18,
18–40, doi: 10.1108/JSIT-06-2015-0059.
[15] S. Shivani, R. Srivastava, N. Tiwari, Developing an e-learning and job portal for IT aspirants, 2022 International
Conference on Applied Artificial Intelligence and Computing (ICAAIC), Salem, India, 2022, 160-165, doi:
10.1109/ICAAIC53929.2022.9792727.
[16] V. S. Tadla, P. M. Singh, K. M. Thakkar, R. Adatkar, Campus placement using machine learning: an extensive
review and comparative study of machine learning methods, 023 6th International Conference on Advances in
Science and Technology (ICAST), Mumbai, India, 2023, pp. 427-430, doi: 10.1109/ICAST59062.2023.10455050.
[17] S. D. Shriramjwar, O. V. Chandure, A study paper on college collaboration portal with training and placement,
International Journal of Research In Science & Engineering, 2018, 3, 78 – 81.
[18] G. Jewani, S. Sahare, T. Kamble, R. Kathalkar, A. Unhale, Online training and placement system, Online training
and placement system, 2023 IEEE International Students' Conference on Electrical, Electronics and Computer
Science (SCEECS), Bhopal, India, 2023, 1-5, doi: 10.1109/SCEECS57921.2023.10063051.
[19] A. S. Sharma, S. Prince, S. Kapoor, K. Kumar, Placement prediction system using logistic regression in logistic
regression, 2014 IEEE International Conference on MOOC, Innovation and Technology in Education (MITE),
Patiala, India, 2014, 337-341, doi: 10.1109/MITE.2014.7020299.
[20] S. Vora, A. Arya, C. Kumbhar, H. Dalvi, An AI-based adaptive assessment system for effective campus placement
process management Available to Purchase, AIP Conference Proceedings, 2023, 2916, 020014,
10.1063/5.0177537.
[21] B. Bhuvaneswaran, R. Reshma, V. Soniya, JobQuench: An intelligent and automated placement management
system for enhanced campus recruitment, 2025 Third International Conference on Augmented Intelligence and
Sustainable Systems (ICAISS), Trichy, India, 2025, 1427-1435, doi: 10.1109/ICAISS61471.2025.11042066.
[22] A. Turkmenbayev, E. Abdykerimova, S. Nurgozhayev, G. Karabassova, D. Baigozhanova, The application of
machine learning in predicting student performance in university engineering programs: a rapid review, Frontiers
in Education, 2025, 10, 1562586, doi: 10.3389/feduc.2025.1562586.
[23] N. Kathirisetty, R. Jadeja, H. K. Thakkar; D. Garg, C. -C. Chang, R. Mahadeva, Student placement probabilistic
assessment using emotional quotient with machine learning: a conceptual case study, IEEE Access, 2023, 11,
125716-125737, doi: 10.1109/ACCESS.2023.3330320.
[24] N. Mezhoudi, R. Alghamdi, R. Aljunaid, G. Krichna, D. Düştegör, Employability prediction: a survey of current
approaches, research challenges and applications, Journal of Ambient Intelligence and Humanized Computing,
2023, 14, 1489–1505, doi: 10.1007/s12652-021-03276-9.
[25] A. Pathak, M. Matcha, M. Gopisetti, S. Joshi, A machine learning framework for predicting student placement
outcomes, 2025, 30, 1715-172, doi: 10.18280/isi.300704.
[26] J. T. Hancock, T. M. Khoshgoftaar, Q. Liang, A problem-agnostic approach to feature selection and analysis using
SHAP, Journal of Big Data, 2025, 12, 2025, doi: 10.1186/s40537-024-01041-1.
[27] J. He, Y. Luo, T. Wang, iDigiChat: intelligent digital marketing service chatbot for providing efficient customer
services using artificial intelligence, Scientific Reports, 2025, 15, 33074, doi: 10.1038/s41598-025-14722-5.
[28] M. L. Bernardi, M. Cimitile, G. Panella, R. Pecori, G. Simoncelli, Automatic generation of job safety reports with
explainable RAG-based LLMs, Information Systems Frontiers, 2025, doi: 10.1007/s10796-025-10634-x.
[29] I. H. Sarker, AI-based modeling: techniques, applications and research issues towards automation, intelligent and
smart systems, SN Computer Science, 2022, 3, 158, doi: 10.1007/s42979-022-01043-x.
[30] H. R. Swapna, D. Arpana, Chatbots as a Game Changer in E-recruitment: An Analysis of Adaptation of Chatbots.
In: Kumar, R., Mishra, B.K., Pattnaik, P.K. (eds) Next Generation of Internet of Things. Lecture Notes in Networks
and Systems, Springer, Singapore. 2021, 201, doi: 10.1007/978-981-16-0666-3_7.
[31] J. He, Y. Luo, T. Wang, iDigiChat: intelligent digital marketing service chatbot for providing efficient customer
services using artificial intelligence, Scientific Reports, 2025, 15, 33074, doi: 10.1038/s41598-025-14722-5.
[32] I. Engeness, M. Nohr, T. Fossland, Investigating AI Chatbots’ role in online learning and digital agency
development, Education Science, 2025, 15, 674, doi: 10.3390/educsci15060674.
[33] Kushal Karwa, Leveraging AI and digital technologies to transform on-campus recruitment for design students:
enhancing employer engagement and hiring outcomes, 2025, 11, doi: 10.22399/ijcesen.3779.
[34] A. Fageeh, The rise of chatbots in higher education: Exploring user profiles, motivations, and integration
strategies, Social Sciences & Humanities Open, 2025, 12, 101996, doi: 10.1016/j.ssaho.2025.101996.
Publisher Note: The views, statements, and data in all publications solely belong to the authors and contributors. GR
Scholastic is not responsible for any injury resulting from the ideas, methods, or products mentioned. GR Scholastic
remains neutral regarding jurisdictional claims in published maps and institutional affiliations.
Open Access
This article is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, which
permits the non-commercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long
as appropriate credit to the original author(s) and the source is given by providing a link to the Creative Commons
License and changes need to be indicated if there are any. The images or other third-party material in this article are
included in the article's Creative Commons License, unless indicated otherwise in a credit line to the material. If
material is not included in the article's Creative Commons License and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view
a copy of this License, visit: https://creativecommons.org/licenses/by-nc/4.0/
© The Author(s) 2026