DOI : 10.17577/IJERTV15IS020430
- Open Access
- Authors : Parth H. Rupareliya, Dr. Bhoomi M. Bangoria
- Paper ID : IJERTV15IS020430
- Volume & Issue : Volume 15, Issue 02 , February – 2026
- Published (First Online): 23-02-2026
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License:
This work is licensed under a Creative Commons Attribution 4.0 International License
Ai-Powered Personalized Career Coach: Experimental Evaluation and Performance Analysis of an Adaptive Career Guidance System
Parth H. Rupareliya
PG Scholar Department of Computer Science and Engineering (Data Science)
Dr. Bhoomi M. Bangoria
Assistant Professor, Department of Information Technology Dr. Subhash University, Junagadh – 362001, Gujarat, India
Abstract:- This study presents the experimental evaluation and performance analysis of a fully implemented AI-powered personalized career coaching system. Building upon the architectural framework proposed in our previous work, this paper demonstrates the practical effectiveness of the system through rigorous quantitative experimentation on a dataset of 185,000+ records. The implemented system integrates a hybrid machine learning pipeline comprising Random Forest, Gradient Boosting, and BERT-based NLP components with a standardized prompt engineering module and a guided React.js user interface. Experimental results demonstrate that the proposed system achieves a career recommendation accuracy of 87.4%, a skill gap identification precision of 91.2%, and a user satisfaction score of 4.3/5.0, outperforming baseline systems including traditional rule-based portals, standalone ML methods, and non-personalized platforms by margins of 23 41%. The system also achieves an average API response time of
1.8 seconds and supports concurrent multi-user load with 94.6% uptime under stress testing. The standardized prompt engineering module achieved 95.3% response consistency across 10,000 test interactions. Results validate the hypothesis that combining ML, NLP, standardized prompt templates, and a guided user interface significantly improves the quality, consistency, and accessibility of career guidance over existing approaches.
Index Terms – Artificial Intelligence, Career Coaching, Machine Learning, Natural Language Processing, Experimental Evaluation, Performance Analysis, Personalized Recommendations, Adaptive Systems, Prompt Engineering
-
INTRODUCTION
Career guidance and planning have traditionally relied on human counselors and static assessment tools that often fail to adapt to rapidly changing job market dynamics and individual user needs [3,5]. The emergence of artificial intelligence technologies presents unprecedented opportunities to revolutionize career coaching through personalized, data-driven approaches that can analyze vast datasets of career paths, job trends, and skill requirements in real-time.
Our previous work [IJARPR0942] proposed a comprehensive architectural framework for an AI-powered personalized career coaching system, detailing the system design, standardized prompt engineering approach, multi-layered technology stack, and dataset sources. That work
demonstrated theoretical promise and reported preliminary system behavior. However, it did not provide quantitative experimental validation of system effectiveness or a rigorous comparative analysis against competing approaches. This paper directly addresses that gap.
The 4th semester work represents the transition from system design and prototyping to full system implementation, deployment, and experimental evaluation. The implemented system processes user inputs including skills, interests, career goals, preferred industries, and educational background through a four-step guided interface and produces personalized career recommendations, skill gap analyses, and learning path suggestions. The core research question addressed in this paper is: How effectively does the proposed AI-powered career coaching system perform relative to existing career guidance approaches in terms of recommendation accuracy, user satisfaction, response consistency, and system efficiency?
Experimental results across five evaluation dimensions recommendation accuracy, skill gap identification, user satisfaction, prompt consistency, and system performance confirm that the proposed system significantly outperforms all baseline systems evaluated. The system achieves a top-3 career recommendation accuracy of 87.4%, reduces the average time for a user to receive actionable career guidance from approximately 45 minutes (human counseling) to 2.3 minutes, and maintains 95.3% output consistency through the standardized prompt module.
The remainder of this paper is structured as follows. Section
II reviews related literature. Section III describes the implemented system. Section IV details the experimental setup. Section V presents and discusses results. Section VI presents a comparative analysis. Section VII discusses limitations and future work. Section VIII concludes the paper.
-
LITERATURE REVIEW
-
AI Applications in Career Planning
Recent research has demonstrated the significant potential of artificial intelligence in enhancing career planning processes. Zhao et al. (2025) developed an AI-based career planning system utilizing machine learning and data mining techniques, achieving improved decision accuracy compared to traditional methods [3]. Their work highlighted the importance of systematic data processing and algorithmic approaches in career guidance, but lacked real-time adaptability and a dynamic feedback loop limitations that the present system directly addresses.
The integration of AI technologies in career planning has shown particular promise in handling large datasets and identifying patterns that may not be apparent through conventional analysis methods [5,15]. Machine learning algorithms can process multiple variables simultaneously, including individual skills, market trends, educational backgrounds, and industry requirements, to generate comprehensive career recommendations.
-
Machine Learning for Recommendation Systems
Chen et al. (2024) applied deep neural networks including LSTM and CNN architectures to career path prediction, achieving notable accuracy improvements over shallow models [7]. However, their work identified a critical limitation: deep learning models suffer from interpretability issues, making it difficult to explain recommendations to end users. The present system addresses this through its standardized prompt engineering module, which generates human-readable reasoning alongside every recommendation.
Kumar et al. (2025) demonstrated real-time adaptive recommendation systems using reinforcement learning techniques [10]. While their approach showed strong adaptation over time, computational complexity for large- scale deployment remained a challenge. The present system's use of ensemble methods balances accuracy with deployment efficiency.
-
Prompt Engineering for AI Systems
Williams (2025) established best practices for prompt engineering in career AI applications, demonstrating that structured prompt templates significantly improve consistency of AI-generated outputs [9]. However, that work identified limited standardization across different AI platforms as a key gap. The standardized prompt module in the present system implements a three-template architecture
Career Exploration, Skill Gap Analysis, and Learning Path prompts specifically designed to address this gap.
-
User Experience in Career Guidance Systems
Patel and Desai (2024) investigated user experience design in AI career systems, finding that guided, step-by-step interfaces significantly improve user task completion rates and satisfaction scores compared to unguided interfaces [8]. Their
study found that users with non-technical backgrounds were
3.2x more likely to complete a career assessment on a guided platform. The four-step guided interface implemented in the present system is directly informed by these findings.
-
Feedback Loops in Adaptive Systems
Garcia et al. (2024) demonstrated that effective feedback loops using active learning and Bayesian optimization improve adaptive AI system performance over time [16]. However, their work noted limited research on long-term feedback integration and model stability a direction pursued in the present system through incremental online learning mechanisms applied to the recommendation engine.
-
-
SYSTEM IMPLEMENTATION
-
Implemented System Architecture
The fully implemented system follows a three-layer architecture described in our prior work, now realized as a working application. The Presentation Layer is built with React.js and HTML5, featuring a four-step guided workflow:
(1) User Profile Creation, (2) Interest and Preference Mapping, (3) Career Goals Definition, and (4) Results and Recommendations Dashboard. The interface employs real- time input validation, progress indicators, and responsive design verified across desktop, tablet, and mobile viewports.
(Fig. 1: Proposed AI-Powered Personalized Career Coach System Architecture (Frontend, Backend, External APIs, and Database Layers)
The Application Layer is implemented in Python using FastAPI for the backend API server, with asynchronous task
handling via Celery. The layer includes the Prompt Standardization Module, which houses three parameterized prompt templates, and the AI/ML Processing Engine comprising three sub-models: the Career Recommendation Model (SPM), the Skill Gap Path Generator, and the Learning Path Boosting module.
(Fig. 2: Implemented System User Interface User Profile Creation Screen displaying Education & Skills input panel with skill tagging, experience level dropdown, target role, and weekly learning hours;)
(Fig. 3: Implemented System User Career Assessment Module showing Step 1 of 4 Work Style Preference Selection with 25% progress)
The Data Layer employs three database systems: MongoDB for user profiles and feedback data (document-oriented storage), PostgreSQL for structured career data and job market records (relational storage), and Redis for caching
and session management (in-memory storage). Security is implemented via SSL/TLS encryption, OAuth 2.0 authentication, and JWT token management.
-
Machine Learning Pipeline
The career recommendation engine employs a Gradient Boosting ensemble (XGBoost) trained on 42,000 labeled career transition records from the Kaggle Job Recommendations dataset. Feature inputs to the model include: current skills (TF-IDF vectorized, 512 dimensions), years of experience (normalized), education level (ordinal encoded), preferred industry (one-hot encoded, 24 categories), and work style preference (categorical, 4 values). The model outputs a ranked list of top-5 career recommendations with associated confidence scores.
The Skill Gap Analysis module uses cosine similarity between the user's current skill vector and the required skill vectors for the target career, derived from the Open Skills Project taxonomy of 8,000+ standardized competencies. Gaps are ranked by importance weight, computed from LinkedIn job posting frequency data.
The NLP module employs a BERT-base transformer fine- tuned on 15,000 career-related text samples extracted from Coursera course descriptions and career guidance documents. This module processes free-text inputs such as user- described career goals and maps them to structured career taxonomy entries.
(Fig. 4: Skill Gap Analysis Interface User inputs target role (Software Engineer) and initiates AI-powered skill comparison against required competencies derived from Open Skills Project taxonomy and LinkedIn job posting data)
-
Standardized Prompt Engineering Module
Three parameterized prompt templates were developed and implemented:
Career Exploration Prompt: "Based on user's skills:
{skills_list}, interests: {interests}, education:
{education_level}, and experience: {years}, recommend top careers with reasoning."
Skill Gap Analysis Prompt: "Compare user's current skills:
{current_skills} with required skills for {target_career}. Identify gaps and prioritize by importance."
Learning Path Prompt: "Create a structured learning path for user to transition from {current_role} to {target_role} considering time: {available_hours} and budget: {budget}."
Each template is validated against a schema before submission to the AI backend, ensuring structural completeness and reducing null or malformed responses. This module achieved 95.3% response consistency in testing (described in Section V).
-
Feedback Loop Implementation
The feedback loop collects both explicit ratings (15 star post-session rating) and implicit signals (recommendation click-through rate, time-on-recommendation-page). These signals are aggregated weekly and used to retrain the career recommendation model incrementally via online learning, without requiring full retraining of the model. Model updates are validated on a held-out evaluation set before deployment.
-
-
EXPERIMENTAL SETUP
-
Dataset
Experiments were conducted using a consolidated dataset drawn from five sources, totaling 185,000+ records.
Data Source
Records
Usage
Kaggle Job Recommendations
50,000+
Training recommendation model
LinkedIn Job Posting Data
100,000+
Market trend validation
Coursera Course Metadata
15,000+
Learning path recommendations
Udemy Learning Data
12,000+
Learning resource optimization
Open Skills Project
8,000+
Skill standardization and mapping
Total
185,000+
Full system training and evaluation
(Table 1: Dataset Sources and Sizes Used in Experimentation)
For model training and evaluation, the Kaggle dataset was split 70% training, 15% validation, and 15% test. The
LinkedIn dataset was used exclusively for market trend validation and was not used in model training to prevent data leakage.
-
Evaluation Metrics
Five evaluation dimensions were defined:
-
Recommendation Accuracy: Top-1, Top-3, and Top-5 accuracy of career recommendations against ground-truth career paths in the held-out test set (7,500 records).
-
Skill Gap Identification Precision and Recall: Precision, Recall, and F1-Score of skill gaps identified by the system compared to expert-labeled skill gap ground truth (500 manually labeled profiles).
-
Prompt Consistency Score: Percentage of 10,000 repeated identical inputs that produce structurally and semantically consistent outputs from the prompt module.
-
User Satisfaction Score: Mean user rating (15 scale) collected from 200 test users who interacted with the full system.
-
System Performance Metrics: Average API response time (milliseconds), throughput (requests/second under load), and uptime percentage under 500-concurrent-user stress test.
-
-
Baseline Systems for Cmparison
The proposed system was compared against four baselines:
-
Baseline 1 Traditional Rule-Based Portal: A keyword-matching career portal using static job-role mappings and fixed skill checklists.
-
Baseline 2 Standalone ML (Random Forest): A Random Forest classifier trained on the same dataset without NLP, prompt engineering, or guided UI.
-
Baseline 3 Zhao et al. (2025) System: The ML + Data Mining system reported in prior literature, replicated using published methodology.
-
Baseline 4 Non-Personalized AI: A general-purpose LLM queried without standardized prompt templates or user profile context.
-
-
RESULTS AND ANALYSIS
-
Career Recommendation Accuracy
This presents recommendation accuracy results across all systems evaluated on the 7,500-record held-out test set.
System
Top-1 Accuracy
Top-3 Accuracy
Top-5 Accuracy
Traditional Rule- Based Portal
41.2%
52.7%
61.4%
Standalone ML (Random Forest)
58.6%
70.3%
78.1%
Zhao et al. (2025) System
61.4%
73.8%
80.2%
Non-Personalized AI (no prompt)
54.3%
67.9%
75.6%
Proposed System (Full)
74.8%
87.4%
93.1%
identifies a skill gap, it is correct in 91 out of 100 cases critical for user trust. The recall of 88.7% indicates that the system correctly identifies 88.7% of all actual skill gaps, minimizing the risk of users missing important development areas.
5.3 Prompt Consistency Analysis
The standardized prompt module was evaluated by submitting 10,000 identical user profile inputs across three prompt templates and measuring structural and semantic consistency of outputs.
(Table 2: Career Recommendation Accuracy Comparison)
The proposed system achieves a Top-3 accuracy of 87.4%, representing a 25.2 percentage point improvement over the best-performing baseline (Zhao et al., 2025 at 73.8%) and a
35.9 percentage point improvement over the traditional rule-based portal. Top-5 accuracy reaches 93.1%, indicating that in over 9 out of 10 cases, the correct career path appears within the system's top-5 recommendations.
The ablation analysis (Table 3) reveals the contribution of each system component to this accuracy improvement.
The ablation study confirms that each component contributes meaningfully to system performance. The guided UI contributes 6.0 percentage points by ensuring cleaner, more complete user inputs. The standardized prompt module contributes 4.3 percentage points through consistent AI interaction. The NLP module contributes 9.8 percentage points by accurately processing free-text user goals.
-
Skill Gap Identification Performance
This presents Precision, Recall, and F1-Score for skill gap identification, evaluated on 500 manually labeled user profiles by two career domain experts (inter-rater agreement = 0.84).
System
Precision
Recall
F1-Score
Rule-Based System
62.3%
54.1%
57.9%
Standalone ML
74.8%
69.2%
71.9%
Zhao et al. (2025)
77.4%
72.6%
74.9%
Proposed System
91.2%
88.7%
89.9%
(Table 3: Skill Gap Identification Performance)
The proposed system achieves a Precision of 91.2% and Recall of 88.7%, yielding an F1-Score of 89.9%. This represents a 15.0 point F1 improvement over the next-best baseline. High precision (91.2%) means that when the system
Prompt Template
Structural Consistency
Semantic Consistency
Overall Consistency
Career Exploration Prompt
97.4%
93.8%
95.6%
Skill Gap Analysis Prompt
98.1%
94.2%
96.2%
Learning Path Prompt
96.7%
92.8%
94.8%
Average (All Templates)
97.4%
93.6%
95.3%
Non- Standardized (Baseline)
73.1%
61.4%
67.3%
(Table 4: Prompt Consistency Results)
The standardized prompt module achieves an overall consistency score of 95.3%, compared to only 67.3% for a non-standardized (free-form) prompting baseline a 28.0 percentage point improvement. This result validates the central claim of the prompt engineering approach: that standardized templates produce predictably consistent outputs suitable for production deployment.
-
Visual Proof of Prompt Consistency
prompt module (proposed system) and a non-standardized baseline approach across three dimensions: structural consistency, semantic consistency, and overall consistency. The proposed standardized prompt module achieves 95.3% overall consistency compared to 67.3% for non-standardized prompts (+28.0pp improvement). Results based on 10,000 identical test inputs across three prompt templates.
-
Users completed identical career guidance tasks on both systems:
-
Create profile with skills, education, experience
-
Complete career assessment
-
Review recommendations
-
Analyze skill gaps for target career
-
-
After using each system, users rated 5 dimensions on a 5- point Likert scale:
-
1 = Very Dissatisfied
2 = Dissatisfied
3 = Neutral
(Fig.5 – Prompt Consistency Comparison Standardized vs. Non-Standardized Prompting)
-
User Satisfaction and Usability
A total of 200 users (120 university students and 80 working professionals, aged 1938) completed a structured system evaluation session. Each user interacted with both the proposed system and a baseline portal, in randomized order, and rated each system across five dimensions on a 5-point Likert scale.
-
User Satisfaction Data Collection and Calculation Methodology
Participant Recruitment: 200 users (120 university students aged 19-26, 80 working
professionals aged 27-38) were recruited through Dr.
Subhash University career services
department and professional networking groups in Gujarat.
Experimental Protocol:
-
Each user was randomly assigned to test either:
-
Group A: Proposed System first, then Rule-Based Portal
-
Group B: Rule-Based Portal first, then Proposed System (Randomization prevents order bias)
-
4 = Satisfied
5 = Very Satisfied
Sample Rating Sheet:
User #47 (Age 24, Final Year B.Tech Student):
Rule-Based Portal Ratings: Ease of Use (3), Relevance (2), Clarity (2), Quality (3), Overall (3)
Proposed System Ratings: Ease of Use (5), Relevance (4), Clarity (5), Quality (4), Overall (5)
Calculation Method:
For each dimensio, mean score = (Sum of all user ratings) / (Number of users)
Example – Ease of Use (Proposed System):
Sum = (5 + 4 + 5 + 4 + … + 5) across 200 users = 900
Mean = 900 / 200 = 4.5
Aggregated Results (Proposed System):
– Ease of Use: 4.5/5.0 (SD = 0.6)
-
Relevance of Recommendations: 4.4/5.0 (SD = 0.7)
-
Clarity of Skill Gap Analysis: 4.3/5.0 (SD = 0.8)
-
Quality of Learning Path: 4.2/5.0 (SD = 0.7)
-
Overall Satisfaction: 4.3/5.0 (SD = 0.7)
Statistical Validation:
Mann-Whitney U test was conducted to compare proposed system vs baselines.
Result: U = 3,847, p < 0.001, effect size r = 0.68 (large effect)
Conclusion: User satisfaction difference is statistically significant.
Qualitative Feedback Themes (from open-ended questions):
-
89% of users praised the "step-by-step guided interface"
-
84% valued the "visual skill gap dashboard"
-
76% appreciated "personalized learning resource recommendations"
-
12% requested additional features (voice interface, multilingual support)
-
-
-
Model Performance Over Feedback Iterations
The feedback loop mechanism was evaluated by simulating 8 weekly retraining cycles using accumulated user interaction data. Figure 1 (described below) shows the improvement in Top-3 recommendation accuracy over successive feedback iterations.
Feedback Iteration
Top-3 Accuracy
Improvement vs. Baseline
Iteration 0 (No Feedback)
82.1%
Iteration 1 (1 week)
83.4%
+1.3%
Iteration 2 (2 weeks)
84.7%
+2.6%
Iteration 4 (4 weeks)
85.9%
+3.8%
Iteration 6 (6 weeks)
86.8%
+4.7%
Iteration 8 (8 weeks)
87.4%
+5.3%
(Table 5: Model Accuracy Improvement Over Feedback Iterations)
The results confirm that the feedback loop mechanism provides meaningful continuous improvement. Over 8 weeks of simulated feedback integration, Top-3 accuracy improved from 82.1% to 87.4% (+5.3 percentage points), validating the adaptive learning design. The rate of improvement shows a concave curve, indicating that gains are largest in the early iterations and stabilize as the model converges consistent with expected online learning behavior [16].
-
Comparative Visual Analysis
To provide a comprehensive view of system performance relative to baselines, this section presents three comparative visualizations across all evaluation dimensions.
-
Multi-Metric Comparison Across All Systems:
-
The comparative analysis presented in Figure 7 reveals that the proposed system consistently outperforms all baseline approaches across every evaluation metric. The proposed system achieves Top-3 recommendation accuracy of 87.4%, representing a 13.6 percentage point improvement over the next-best system (Zhao et al., 2025 at 73.8%) and a 34.7 percentage point improvement over the rule-based portal baseline (52.7%).
In skill gap identification, the proposed system's F1-Score of 89.9% exceeds the best baseline (Zhao et al., 74.9%) by 15.0 percentage points, demonstrating superior capability in accurately identifying missing competencies for target career transitions. The prompt consistency metric shows the most dramatic improvement, with the proposed system achieving 95.3% consistency compared to 67.3% for the non- standardized AI baselinea 28.0 percentage point gain that validates the standardized prompt engineering approach as a critical architectural component rather than merely an implementation detail. User satisfaction results further confirm the system's practical effectiveness, with an average rating of 4.3/5.0 (86% normalized) compared to 2.7/5.0 (54%) for the rule-based portal and 3.1/5.0 (62%) for the non- personalized AI system.
Notably, no baseline system achieves above-average performance across all four metrics simultaneously.
(Fig. 6: Comprehensive System Performance Comparison Across Four Key Metrics)
-
Key Findings
-
-
DISCUSSION
engineering module, and a guided four-step React.js user interface. Rigorous evaluation across five performance dimensions conducted on a 185,000+ record dataset with
200 human participants produced the following key
The experimental results validate three core hypotheses of this research. First, the standardized prompt engineering module demonstrably improves AI output consistency (95.3% vs. 67.3% for unstructured prompting), confirming that prompt standardization is a meaningful engineering contribution rather than merely an implementation detail. Second, the combination of ML, NLP, guided UI, and standardized prompts produces an accuracy (87.4% Top-3) that meaningfully exceeds any individual component or baseline system, confirming the value of the integrated multi- component architecture. Third, the guided four-step UI significantly improves task completion rate (94.5% vs. 71.2%) and user satisfaction, supporting the hypothesis that user interface design is a first-class engineering concern in career AI systems.
-
Limitations
Several limitations of the current system should be acknowledged. The training data for the recommendation model is predominantly drawn from English-language job market datasets, limiting applicability to non-English- speaking regions. The user evaluation study (n=200) was conducted primarily with university students and young professionals in India, which may limit generalizability across demographics. The feedback loop simulation used synthetic interaction data for iterations beyond the first two weeks; real-world feedback collection over a longer period may show different improvement trajectories. Additionally, while the system demonstrates strong performance on the Kaggle-derived test set, performance on entirely novel career domains not well-represented in the training data may be lower.
-
Future Work
Future research directions include: (1) expansion of training datasets to include regional Indian job market data (NASSCOM, Naukri.com) and non-English career domains;
(2) integration of psychometric assessment features to improve recommendation personalization beyond skill-based matching; (3) development of a voice-based interaction interface to improve accessibility; (4) longitudinal evaluation of real-world career outcome tracking to validate whether system recommendations translate to successful career transitions; and (5) deployment of the system within a university career services context for a controlled long-term user study.
-
-
CONCLUSION
This paper presented the experimental evaluation of a fully implemented AI-powered personalized career coaching system. The system integrates a hybrid ML pipeline (XGBoost + BERT), a standardized three-template prompt
results: Top-3 career recommendation accuracy of 87.4% (outperforming all baselines by 13.634.7 percentage points), skill gap identification F1-Score of 89.9%, prompt consistency of 95.3%, overall user satisfaction of 4.3/5.0, task completion rate of 94.5%, average response time of 1.8 seconds at 100 concurrent users, and a 19.5x reduction in time-to-guidance versus human counseling at ~0.03% of the cost.
<>The ablation study confirmed that every architectural component ML engine, NLP module, standardized prompts, and guided UI contributes meaningfully to overall system performance. The feedback loop mechanism demonstrated continuous improvement of 5.3 percentage points in Top-3 accuracy over 8 simulated retraining cycles. These results collectively validate the proposed system as an effective, scalable, and cost-efficient alternative to both traditional career portals and human career counseling, with particular strength in personalization, consistency, and accessibility.
The system represents a significant step toward the democratization of high-quality career guidance, making personalized, data-driven career coaching accessible to users regardless of their geographic location, economic resources, or technical background.
ACKNOWLEDGMENTS
The authors acknowledge the support and guidance provided by the faculty and administration of Dr. Subhash University, Junagadh, in facilitating this research work. Special recognition is extended to the Department of Computer Science and Engineering for providing the necessary resources and academic environment conducive to research and development activities.
Gratitude is expressed to the various data providers and open- source communities whose datasets and tools contributed to the successful development and validation of the proposed system. The collaborative nature of modern AI research is exemplified by these contributions to the broader scientific community.
REFERENCES
-
Kasem, M.S., et al. (2024). "Customer profiling, segmentation, and sales prediction using AI in direct marketing." Springer.
-
Pitka, T., et al. (2024). "Time analysis of online consumer behavior by decision trees, GUHA association rules, and formal concept analysis." Springer.
-
Zhao, L. (2025). "AI-Based Intelligent Career Planning System."
Proceedings of ICAAAI, Atlantis Press.
-
Adhikari, S. (2024). "Gender Differences in Career Coaching Outcomes: A Quantitative Analysis Using a Logit Model." ResearchGate.
-
Kumar, M., & Sharma, R. (2023). "AI and ML in Personalized Education and Career Recommendations." International Journal of Advanced Computer Science, 13(4), 112120.
-
Singh, A. (2023). "AI-Driven Guidance for Career Development."
Journal of Emerging Technologies in Learning, 18(2), 7585.
-
Chen, L., et al. (2024). "Deep Learning for Career Path Prediction."
IEEE Transactions on Learning Technologies.
-
Patel, R., & Desai, K. (2024). "User Experience Design in AI Career Systems." ACM CHI Conference Proceedings.
-
Williams, J. (2025). "Prompt Engineering Best Practices for Career AI." arXiv preprint.
-
Kumar, S., et al. (2025). "Real-time Adaptive Recommendation Systems." ICML Proceedings.
-
Zhang, Y., et al. (2025). "Reinforcement Learning for Personalized Learning Paths." NeurIPS.
-
Brown, M., & Taylor, S. (2024). "Bias Detection in Career Recommendation Systems." FAccT Conference.
-
Anderson, K. (2024). "Scalable AI Architectures for EdTech." IEEE Cloud Computing, 11(3).
-
Lee, J., & Park, H. (2024). "Natural Language Processing for Career Guidance." EMNLP.
-
Thompson, R. (2024). "Evaluation Metrics for Career AI Systems."
Journal of AI Research, 42.
-
Garcia, M., et al. (2024). "Feedback Loops in Adaptive AI Systems."
AAMAS Conference.
-
Kumar, R., et al. (2023). "A systematic review on big data applications and scope for industrial processing & healthcare sectors." Springer.
-
Rupareliya, P.H., & Bangoria, B.M. (2025). "AI-Powered Personalized Career Coach: A Comprehensive Approach to Adaptive Career Guidance Systems." International Journal of Advance Research Publication and Reviews, Vol. 02, Issue 09, pp. 454463.
-
Kaggle Job Recommendations Dataset. Available at: https://www.kaggle.com/datasets/
-
LinkedIn Economic Graph and Job Posting Data. Available at: https://economicgraph.linkedin.com/
-
Coursera Course Metadata API. Available at: https://www.coursera.org/
-
Open Skills Project Database. Available at: https://www.openskillsnetwork.org/
