About WhyWrong-AI
The Inspiration
Every student has experienced it: you submit an answer, get it wrong, see the correct solution, and move on. But one important question often remains unanswered: Why was my answer wrong?
A student might repeatedly make the same mistake in mathematics, misunderstand a physics principle, or apply an incorrect logic pattern in computing. Traditional assessment systems can identify incorrect answers, but they do not always uncover the misunderstanding behind them.
This inspired me to create WhyWrong-AI Fix the Thinking, Not Just the Answer.
I wanted to build an AI-powered learning assistant that goes beyond marking answers right or wrong. Instead, WhyWrong-AI investigates the learner's reasoning, identifies possible misconceptions, tests those assumptions, and provides personalized guidance to help students understand concepts more deeply.
What I Built
WhyWrong is an AI-powered educational platform designed to diagnose misconceptions rather than simply grade answers.
The platform follows an intelligent learning process:
- Diagnose: The AI examines a student's answer and reasoning to identify possible misconceptions.
- Probe: It generates adaptive follow-up questions to test whether its initial diagnosis is correct.
- Remediate: It creates personalized explanations, examples, and practice worksheets based on the identified learning gap.
- Re-test: It schedules follow-up practice to determine whether the student has overcome the misconception.
- Report: It provides teachers with insights into individual learning difficulties and common misconceptions across their classes.
WhyWrong-AI is designed around three initial subject areas: Mathematics, Physics, and Computing. Additional features include an interactive concept graph, a Counterexample Lab, an AI decision timeline, teacher review and approval controls, printable PDF reports, and support for voice, camera, and multilingual learning interactions.
An important part of the system is that AI-generated diagnoses are treated as hypotheses supported by evidence, not unquestionable facts. Teachers can review, accept, edit, or reject the AI's conclusions. When the system encounters an unfamiliar misconception, it can propose it as an unverified possibility rather than presenting it as established knowledge.
How I Built It
I approached WhyWrong-AI as an AI-agent-based learning platform rather than a conventional quiz application.
The architecture separates the user interface, backend logic, database, and AI services to make the system modular and easier to extend.
The frontend uses React, Vite, TypeScript, and Tailwind CSS to deliver a responsive and interactive learning experience. The backend is built around PHP, with MySQL for managing learners, questions, assessment attempts, diagnosed misconceptions, remediation activities, teacher reviews, and learning progress.
The AI component is designed to analyse student responses, generate diagnostic follow-up questions, explain identified misconceptions, and produce targeted remediation materials. The agent's decisions and supporting evidence are made visible through an activity timeline.
I also considered practical deployment constraints, designing the application around technologies that can run on affordable shared hosting rather than depending entirely on expensive infrastructure.
What I Learned
Building WhyWrong-AI strengthened my understanding of how AI agents can be applied to real educational problems.
Some of the most important lessons were:
AI should investigate, not just generate: A useful educational AI system needs to ask questions, evaluate evidence, and revise its conclusions. A wrong answer contains useful information: Errors can reveal how a learner understands a concept and where their reasoning breaks down. Human oversight matters: Teachers should remain involved in reviewing important AI-generated educational decisions. Personalization requires context: Effective remediation should address the specific misconception instead of giving every student the same explanation. Trust is essential: AI conclusions should communicate their evidence and uncertainty rather than pretending to be certain.
Challenges I Faced
One of the biggest challenges was designing a system that could distinguish between a simple calculation mistake, a misunderstanding of a concept, and an incorrect reasoning process. A single wrong answer is not always enough to establish the cause of an error.
Another challenge was coordinating multiple AI-agent activities into one understandable learning process. Diagnosis, adaptive questioning, remediation, and re-testing need to work together instead of behaving like disconnected features.
I also had to consider the practical limitations of AI API costs, hosting resources, and deployment. This influenced my decision to use a modular architecture and consider fallback mechanisms so the application can remain demonstrable even when an external AI service is unavailable.
Finally, balancing an advanced AI experience with a simple interface was important. The platform needed to be powerful enough for teachers while remaining understandable and welcoming to students.
The Vision
I believe education should not stop at telling students that they are wrong. It should help them understand why they are wrong and guide them toward getting it right.
WhyWrong-AI aims to make learning more personal, evidence-driven, and meaningful by helping educators discover not only what students do not understand, but also why.
WhyWrong-AI, Fix the Thinking, Not Just the Answer.
Built With
- adaptive
- ai
- css
- edtech
- educational
- generative
- language
- learning
- mysql
- php
- processing
- react
- software
- tailwind
- technology
- typescript
- visualization
- vite