QGenAI: A Web-Based System for Competitive Examination Practice, Difficulty-Stratified Analytics, and Mistake-Driven AI Tutoring

Abstract

Preparing for national competitive examinations such as the Joint Entrance Examination (JEE), the National Eligibility cum Entrance Test (NEET), and the Graduate Aptitude Test in Engineering (GATE) requires regular multiple-choice question (MCQ) practice, reliable answer verification, and systematic mistake revision. When aspirants rely on disconnected textbooks, static question portals, and standalone chat tools, practice records become fragmented and conversational assistance lacks awareness of a learner’s actual errors. This paper presents QGenAI, an authenticated web-based practice, difficulty analysis, and personalized tutoring platform built for JEE, NEET, and GATE preparation. Learners configure custom practice sessions by selecting an examination profile, subject, topic, difficulty tier (Easy, Medium, or Hard), and question count (1, 5, 10, 15, or 20). The application retrieves active MCQs from a structured question bank while withholding answer keys from client payloads to preserve assessment integrity. Submitted responses are graded server-side, persisted with idempotent retry logic, and aggregated into account-scoped analytics covering session completion, difficulty-wise accuracy, seven-day activity, and topic-level mistake queues. Furthermore, an integrated AI Tutor module connects to external large language model (LLM) providers (Groq, Gemini, OpenAI, or Mistral) to deliver both general topic lessons and targeted remediation grounded in up to five of the learner’s most recent incorrect attempts. The system is implemented using React, Vite, Python FastAPI, and Supabase PostgreSQL with Row-Level Security (RLS). We detail the four-tier architecture, relational schema, REST interface, evaluation formulation, and a worked empirical example across 28 completed sessions.

Introduction

1. The QGenAI – Intelligent Competitive Exam Question Generation, Difficulty Analysis and Personalized Practice System is a web-based application designed to help students prepare for competitive examinations such as JEE, NEET, and GATE. Traditional exam preparation methods involve using printed question banks, online test platforms, spreadsheets, and separate AI tools. These methods make it difficult for students to track their performance, identify weak topics, and receive personalized explanations for incorrect answers.

   The proposed QGenAI system combines multiple-choice question (MCQ) practice, difficulty-based performance analysis, and AI-powered personalized tutoring into a single platform. Students can select their examination, subject, topic, and difficulty level (Easy, Medium, or Hard) and practice questions from a curated question database.

   The system automatically evaluates submitted answers and stores students' performance records. It provides detailed analytics, including overall accuracy, difficulty-wise performance, topic-wise progress, and seven-day activity. Based on the student's incorrect answers, the AI Tutor provides personalized explanations and targeted learning suggestions to help improve conceptual understanding.

   The application follows a four-tier architecture consisting of a React frontend, Python FastAPI backend, Supabase PostgreSQL database, and external AI language model services. It supports AI integration with Groq, Google Gemini, OpenAI, and Mistral. Security features such as JWT authentication, Row-Level Security (RLS), and server-side answer validation protect student information and examination content.

   The main aim of QGenAI is to make competitive exam preparation more organized, interactive, personalized, and effective through intelligent practice and performance tracking.

   Main Objectives

   1. To develop a web-based competitive examination preparation platform supporting JEE, NEET, and GATE.

   2. To provide multiple-choice questions at Easy, Medium, and Hard difficulty levels.

   3. To automatically evaluate answers and calculate student performance.

   4. To identify weak subjects and topics through performance analytics.

   5. To provide personalized AI-based explanations for incorrectly answered questions.

   6. To maintain individual student practice history and progress records.

   7. To ensure secure authentication and prevent unauthorized access to answers and personal records.

   Existing System

   Traditional competitive exam preparation methods rely on printed question banks, separate online test series, manual spreadsheets, and general-purpose AI chatbots. Students must manually track their scores and identify weak topics. These tools are not integrated, making personalized learning difficult. General AI chatbots also cannot automatically access students' actual incorrect answers during practice, which limits their ability to provide targeted explanations.

   Proposed System

   The proposed QGenAI system integrates question practice, automatic evaluation, performance analytics, and AI tutoring into one platform.

   Students can select their examination, subject, topic, and difficulty level to practice MCQs. Questions are retrieved from a verified database to maintain accuracy. After submitting answers, the system calculates scores, stores attempts, and displays performance analytics.

   The AI Tutor analyzes recent incorrect attempts and provides topic-related explanations and recommendations. This helps students understand their mistakes and improve their preparation.

   Methodology and Implementation

   The system consists of the following major modules:

   1. User Authentication: Provides secure registration, login, and password recovery using Supabase Authentication.

   2. Exam and Subject Selection: Supports JEE, NEET, and GATE with their respective subjects and syllabus topics.

   3. MCQ Generation and Practice: Retrieves verified questions from the database based on the selected difficulty level.

   4. Answer Evaluation: Checks submitted answers on the backend without exposing correct answers before submission.

   5. Performance Analysis: Calculates overall accuracy, difficulty-wise performance, topic-wise progress, and seven-day activity.

   6. AI Tutor Assistance: Uses AI models to explain incorrect answers and provide personalized learning guidance.

   7. Practice History: Stores previous practice sessions and student attempts for future reference.

   8. Security and Data Management: Uses JWT authentication, database access restrictions, and Row-Level Security to protect user information.

   Technology Stack

   Component	Technologies
   Frontend	React 18, JavaScript, HTML, CSS
   Development Tool	Vite
   Backend	Python, FastAPI
   Server	Uvicorn
   Database	Supabase PostgreSQL
   Authentication	Supabase Auth, JWT
   Data Validation	Pyd

    

ChatGPT is AI and can make mistakes.

Conclusion

We presented QGenAI, a full-stack web platform that unifies competitive examination MCQ practice, server-side answer evaluation, account-scoped learning analytics, and mistake-driven AI tutoring for JEE, NEET, and GATE aspirants. By decoupling question delivery from server-side grading, enforcing PostgreSQL Row-Level Security, and grounding LLM tutoring prompts in a learner’s actual recorded errors, the system overcomes the fragmentation of conventional study workflows. The worked evaluation across 28 sessions demonstrates how difficulty breakdowns and early-signal mistake queues guide targeted revision, establishing a solid foundation for future adaptive testing and RAG-based question generation.

References

[1] F. O. Egara and C. Mosia, “Effectiveness of AI-based tutoring and assessment systems in mathematics education: A systematic review,” Discover Educ., vol. 5, 2026. [2] X. Wang, Y. Maeda, and H.-H. Chang, “Development and techniques in learner model in adaptive e-learning system: A systematic review,” Comput. Educ., vol. 225, Art. no. 105184, Feb. 2025. [3] M. Al-Emran and S. I. Malik, “A review on the use of large language models as virtual tutors,” Sci. Educ., vol. 34, no. 2, pp. 877–892, Apr. 2025. [4] H. Bastani, O. Bastani, A. Sungu, H. Ge, Ö. Kabakc?, and R. Mariman, “Generative AI without guardrails can harm learning: Evidence from high school mathematics,” Proc. Nat. Acad. Sci. USA, vol. 122, no. 26, Art. no. e2422633122, Jul. 2025. [5] Z. Shou, M. Xie, J. Mo, and H. Zhang, “Predicting student performance in online learning: A multidimensional time-series data analysis approach,” Appl. Sci., vol. 14, no. 6, Art. no. 2522, Mar. 2024. [6] N. Daheim, J. Macina, M. Kapur, I. Gurevych, and M. Sachan, “Stepwise verification and remediation of student reasoning errors with large language model tutors,” in Proc. Conf. Empirical Methods Natural Lang. Process. (EMNLP), Miami, FL, USA, Nov. 2024, pp. 8386–8411. [7] Z. A. Pardos and S. Bhandari, “ChatGPT-generated help produces learning gains equivalent to human tutor-authored help on mathematics skills,” PLOS ONE, vol. 19, no. 5, Art. no. e0304013, May 2024. [8] E. Kasneci et al., “ChatGPT for good? On opportunities and challenges of large language models for education,” Learn. Individ. Differ., vol. 103, Art. no. 102274, Apr. 2023. [9] M. Riestra-González, M. del P. Paule-Ruíz, and F. Ortin, “Massive LMS log data analysis for the early prediction of course-agnostic student performance,” Comput. Educ., vol. 163, Art. no. 104108, Apr. 2021. [10] S. Wollny, J. Schneider, D. Di Mitri, J. Weidlich, M. Rittberger, and H. Drachsler, “Are we there yet? A systematic literature review on chatbots in education,” Front. Artif. Intell., vol. 4, Art. no. 654924, Jul. 2021. [11] C. Yang, L. Luo, M. A. Vadillo, R. Yu, and D. R. Shanks, “Testing (quizzing) boosts classroom learning: A systematic and meta-analytic review,” Psychol. Bull., vol. 147, no. 4, pp. 399–435, Apr. 2021. [12] D. Ifenthaler and J. Y.-K. Yau, “Utilising learning analytics to support study success in higher education: A systematic review,” Educ. Technol. Res. Dev., vol. 68, no. 4, pp. 1961–1990, Aug. 2020. [13] G. Akçap?nar, A. Altun, and P. A?kar, “Using learning analytics to develop early-warning system for at-risk students,” Int. J. Educ. Technol. High. Educ., vol. 16, Art. no. 40, Oct. 2019. [14] O. Viberg, M. Hatakka, O. Bälter, and A. Mavroudi, “The current landscape of learning analytics in higher education,” Comput. Hum. Behav., vol. 89, pp. 98–110, Dec. 2018. [15] O. O. Adesope, D. A. Trevisan, and N. Sundararajan, “Rethinking the use of tests: A meta-analysis of practice testing,” Rev. Educ. Res., vol. 87, no. 3, pp. 659–701, Jun. 2017.

Copyright

Copyright © 2026 Abishek M P, Dinesh E, Gowtham S, Mohana Preetha S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.