Quizerai

Inspiration

What it does

How we built it

Challenges we ran into

Inspiration

During my JEE preparation, I was overwhelmed by the massive volume of study material such as PDFs, handwritten notes, coaching books, YouTube lectures, and chapter-wise resources.
The biggest challenge was finding high-quality, topic-specific questions from this unstructured content.

Searching for practice questions on a single concept often took hours. Resources were scattered across platforms, passive reading did not help with retention, and manually creating quizzes was unrealistic under time pressure.

This led to a simple but powerful question:

What if an AI could understand any learning material — text, images, PDFs, or videos — and instantly generate accurate, topic-wise quizzes and revision content?

That idea became QuizerAI — an AI-powered platform that transforms any educational content into personalized quizzes, summaries, and adaptive learning experiences within seconds.

What It Does

QuizerAI is an AI-powered learning and assessment ecosystem that unifies students, teachers, and institutions on a single intelligent platform.

The platform is built by leveraging ERNIE 5 / ERNIE 4.5 for deep language understanding and reasoning, and PaddleOCR-VL for accurate document and handwritten text extraction.

For Students

Convert PDFs, images, handwritten notes, and YouTube videos into instant topic-wise quizzes
Generate summaries, flashcards, and structured revision material
Practice CBT-style exams for JEE, NEET, UPSC, CUET, SSC, and board exams
Get instant explanations and step-by-step reasoning via a 24/7 AI Tutor powered by ERNIE

For Teachers

Create virtual classrooms using join codes or QR codes
Auto-generate quizzes, assignments, and homework directly from lesson content
View real-time class analytics such as accuracy, weak topics, and completion rates
Identify struggling students without manual evaluation

For Institutions

Automated attendance through quiz participation
Centralized dashboards for admins and departments
Multi-class and multi-teacher management

QuizerAI turns isolated learning into a connected, collaborative, and data-driven education ecosystem.

How We Built It (ERNIE + PaddlePaddle Focus)

Multimodal AI Pipeline

PaddleOCR-VL is used to extract text, layout, and structure from PDFs, scanned pages, and handwritten notes
Extracted content is converted into Markdown-ready structured text
ERNIE 5 / ERNIE 4.5 models generate:
- Topic-wise questions
- Difficulty-tagged quizzes
- Concept summaries
- Step-by-step explanations

Adaptive Intelligence

ERNIE is fine-tuned (planned) using LLaMA-Factory / Unsloth for:
- Educational reasoning
- Exam-oriented question generation
- Teacher-style explanations

Frontend

Next.js, React.js, Tailwind CSS, Shadcn UI, TypeScript

Backend

FastAPI (Python), async workers for quiz generation
Secure APIs for ERNIE inference and OCR pipelines

Infrastructure

Dockerized services
Scalable cloud deployment
CI/CD using GitHub Actions

Challenges We Faced

Achieving high OCR accuracy for handwritten notes using PaddleOCR-VL
Preserving document structure (tables, formulas, headings) during OCR-to-Markdown conversion
Designing prompts to ensure ERNIE generates exam-relevant and concept-accurate questions
Scaling quiz generation for thousands of concurrent users
Building a complete AI system while managing college academics

Each challenge pushed us to optimize model usage, prompt design, and system architecture.

Accomplishments We’re Proud Of

Built a full multimodal AI application using OCR + LLM reasoning
Successfully converted unstructured PDFs and handwritten notes into quizzes
Generated thousands of topic-wise quizzes automatically
Designed a real-world educational application with high practical impact
Created a scalable and extensible architecture ready for fine-tuning and deployment
Received strong validation from students preparing for competitive exams

What We Learned

OCR quality directly impacts downstream LLM reasoning quality
ERNIE models excel at structured reasoning when paired with clean OCR outputs
Educational AI must focus on concept clarity, not just text generation
Multimodal pipelines unlock far greater real-world value than text-only systems

What’s Next for QuizerAI (Hackathon-Aligned Roadmap)

ERNIE Fine-Tuning

Fine-tune ERNIE 4.5 using Unsloth / LLaMA-Factory for:
- Exam-specific question generation
- Teacher-style explanations
- Difficulty calibration

Advanced PaddleOCR Enhancements

Fine-tune PaddleOCR-VL for:
- Indian handwritten notes
- Low-quality scanned documents
- Mathematical symbols and diagrams

Multimodal Agent System

Build an AI Agent that:
- Reads documents using PaddleOCR
- Reasons using ERNIE
- Adapts quizzes based on learner performance

Adaptive Learning Engine

Personalized learning paths using student performance data
Continuous difficulty adjustment using ERNIE reasoning

Edge & Offline Expansion

Deploy lightweight OCR + inference for offline or low-connectivity regions
Future integration with edge devices for classrooms

Broader Impact

Exam-specific modules for CBSE, ICSE, JEE, NEET, UPSC, SSC
Nationwide rollout across schools and coaching institutes
Multilingual learning using ERNIE’s language capabilities

QuizerAI’s vision
Build an intelligent, multimodal, and adaptive education platform that uses ERNIE and PaddleOCR to make learning active, fair, personalized, and accessible — at scale.