AI-Powered Exam Paper Evaluator with Intelligent Feedback

Revolutionizing education assessment with AI-powered handwritten O-level/A-level/IELTS exam paper evaluator that delivers grades and intelligent feedback in minutes, not hours

Comment

INSPIRATION

The inspiration for the AI Exam Evaluator came from witnessing the overwhelming burden on students facing biased and unfair grading—and on educators, who spend countless hours manually grading exam papers. During conversations with students and professors, I consistently heard the same issues: evaluation consistency, biased grading, and a lack of detailed feedback—different graders might score the same paper differently, and producing comprehensive comments is time-consuming. That’s why I built an AI‑powered solution to keep grading consistent and unbiased while delivering detailed, constructive feedback that helps students improve—and gives educators their time back.

WHAT IT DOES

AI Exam Paper Evaluator

• Accepts scanned PDF versions of O-level, A-level, and IELTS handwritten exam scripts.

• Extracts questions and handwritten answers from multi-page PDFs and divides them into individual cells.

• The system provides grades and intelligent feedback for each response.

• Flag identifies blank answers and presents the final output for human review. (Human-on-the-Loop Approach)

• Generates and stores reports, marks, feedback, and overall grade.

Question Paper Generation

• Generates Cambridge IGCSE 0653 Science Combined questions (MCQ, short/long answer, numerical, and diagram-based questions)

• The user can change difficulty and topic coverage using settings.

• It follows the Cambridge IGCSE 0653 Science Combined question paper template, and the user can export the generated questions to PDF format.

HOW WE BUILT IT

We built it with the assistance of Kiro; we are also using the Gemini API for the LLM, Supabase for the backend, Netlify for hosting, and Namecheap's edu program domain.

• Frontend: React 18, TypeScript, Tailwind CSS

• OCR & Content Recognition: Preprocesses PDF to understand the structure and has OCR to extract the text from the documents; also, our system maps the layout to ensure each cell is linked to the correct subquestion and related handwritten answers.

• AI Core: Using the Gemini 2.0 flash large language model by calling it to check the paper, giving Mark an overall grade, and providing intelligent feedback.

• Backend: Supabase for authentication, database, storage, and edge functions.

• Trust & Safety: Keeping a human-in-the-loop approach for final report approvals.

• Quality Assurance: Agent Hooks automatically updates the document, testing, and vulnerability checking. We also configured the MCP Server functionality to turn draft code into working software free from bugs and performance issues.

CHALLENGES WE RAN INTO

• OCR variance & layout complexity: In the real exam answer script, the PDF has mixed-quality scans due to phone camera quality, rotated pages, and dense multi-column layouts that required a hybrid of text extraction.

• Consistency: Different subjects and marking schemes demanded prompt engineering plus rule-based normalization to keep partial credit consistent without overfitting to one exam style.

• Reliability: Faculty wants advanced reasoning & students need transparent and unbiased feedback.

• Validity: To ensure we cover all topics, difficulty levels, and cognitive skills, we needed a validator after generating the content and a way to adjust

ACCOMPLISHMENTS THAT WE’RE PROUD OF

• For the first time, we are using a human-in-the-loop approach, allowing humans to work alongside AI instead of replacing them entirely.

• We are not only providing marks but also giving an evaluation of whether it is a correct answer or not based on the student's written answer and the justification of marks that increase trust in the result.

• Teacher-quality question sets: High-quality question generation with distractor rationales and marking schemes that read like what an experienced educator would write.

• Our system reliably handles messy, real exam answer scripts (collected from local coaching centers) & faded PDF scans.

• Our system can detect diagrams and give marks for that type of answer.

WHAT WE LEARNED

• Vibe coding, using AI-powered IDE, prompt engineering, configuring MCP servers, managing documentation using hook, agent workflow, human-on-the-loop, using API, and cloud service.

• Mark schemes are essential for real exam script marking and grading.

• The human-on-the-loop approach dramatically improves our system reliability and faculty members' adoption.

• Feedback is necessary for the students so that they can understand their weak points and specifically target those parts to improve their learning.

• Minimizing bias and unfair grading isn't easy. That required consistency in grading, leaving little room for such dishonesty. Also, the generation gap between faculty members and students is also challenging and makes it harder for faculty members to use our system because they think AI will replace them.

• It is quite hard to fully mimic human-acquired abilities (abilities that come through experience) in our system.

WHAT’S NEXT FOR AI-POWERED EXAM PAPER EVALUATOR WITH INTELLIGENT FEEDBACK & QUESTION PAPER GENERATION

Phase 1 (Current)

✅ AI Exam Evaluation System

✅ Cambridge IGCSE Question Paper Generator

✅ Professional PDF Output

✅ User Authentication

Phase 2 (Planned)

🔄 Mark Scheme Generation

🔄 Multi-language Support

🔄 Advanced Analytics Dashboard

🔄 Collaborative Features

Phase 3 (Future)

📋 Additional Exam Boards & University Level Exams Evaluation

🤖 Advanced AI Tutoring

📱 Mobile Application

🌐 API for Third-party Integration

HOW TO TEST THE APP

Step 1: Go to aiexamevaluator.me

Step 2: Create an account using your Gmail

Step 3: Wait for the confirmation email and click on "Confirm Your Email." (If you don't get the email within 3 minutes, check your Spam folder; if it's not there, retry.)

Step 4: Again, go to the Sign In Window, and use that email and password to log in. Please be aware that logging in is not possible until you have confirmed your email.

Step 5: You will receive five evaluations instantly upon logging in.

Step 6: You can either download any handwritten O-level or A-level exam paper script from the internet that includes the mark scheme, solve a paper independently, or use our sample paper with the mark scheme to test our software. We have uploaded a sample A-Level Physics exam script that includes the mark scheme.

Here is the Google Drive Link: https://drive.google.com/drive/folders/1QkXVdZxargr_97WdXod-2YTfAXz_f-0h?usp=sharing

Step 7: Fill in the information carefully. Upload the student paper in the "Upload Student's Exam Paper" section, and upload the corresponding mark scheme in the "Upload Mark Scheme (Optional)" section. (Note: Upload our Physics Solved Paper for testing in the "Upload Student's Exam Paper" section and Mark Scheme A Level Physics in the "Upload Mark Scheme (Optional)" section. Also note that it will work without a mark scheme: to prove that we have uploaded a biology paper, upload it to the "Upload Student's Exam Paper" section; OR, feel free to upload your paper with a mark scheme.

Finally, after getting your report, please leave some feedback so that I can improve my software.

Built With

Share this project:

Updates