Spaces:
Sleeping
Sleeping
|
Download Backend/README.md from smit9604/ExamCraftAI: direct link, hf CLI and curl.
- Browser
- Download file 3.84 kB
-
https://huggingface.co/spaces/smit9604/ExamCraftAI/resolve/main/Backend/README.md
- Command line
-
hf download hf://spaces/smit9604/ExamCraftAI/Backend/README.md
-
curl -L -o README.md https://huggingface.co/spaces/smit9604/ExamCraftAI/resolve/main/Backend/README.md
3.84 kB
ExamCraft.ai - AI-Powered Question Paper Generator
ExamCraft.ai is a high-performance web application designed for educators to generate structured, academically rigorous question papers in seconds. By leveraging Llama 3.3 (70B) via Groq AI, it maps syllabus content directly to Bloom’s Taxonomy cognitive levels.
🚀 Tech Stack & Tools
1. Backend (The Brain)
- Python & Flask: The core web framework handling routing, file management, and API orchestration.
- Groq AI SDK: Used to interface with the Llama 3.3 70B model for lightning-fast inference (low latency).
- PyPDF2 & python-docx: Robust libraries used to parse text from complex PDF and Word documents.
- ReportLab: A professional PDF toolkit used to programmatically generate the final exam papers.
2. Frontend (The Experience)
- Tailored CSS & Bootstrap 5: Implements a "Premium Glassmorphism" design system with mesh gradients and fluid layout.
- Vanilla JavaScript: Handles asynchronous file uploads (AJAX), multi-step state management, and real-time UI updates (like the stepper and preview).
- FontAwesome & Google Fonts: Integrated Plus Jakarta Sans and Outfit for clean academic typography.
🧠 Core Functionalities & Logic
A. Intelligent Text Extraction (extractor.py)
Logic: The system detects the file extension and routes it to the specific parser.
- PDF: Uses
PdfReaderto iterate through pages and extract text layers. - DOCX: Iterates through paragraphs and tables to preserve content order.
- Implementation: Text is stripped of excess whitespace and truncated to fit within optimal AI context windows (Token optimization).
B. AI Question Generation (ai_generator.py)
Logic: We use a "Structured Prompting" strategy. Instead of just asking for questions, the system passes:
- The Context: Extracted text from the syllabus.
- The Cognitive Framework: The selected Bloom’s levels (Remember, Apply, Analyze, etc.).
- The Constraint Matrix: Difficulty level and the user-defined Exam Pattern (marks, question counts).
- The Output Format: Forces the AI into JSON Mode, ensuring the backend can always parse the result into a preview.
C. The 3-Step Workflow Implementation
- Upload Phase: The file is stored with a unique UUID to prevent collisions. JavaScript triggers the extraction instantly.
- Configuration Phase: Users define the exam "Schema." UI buttons dynamically add or remove sections ($N$ questions of $X$ marks).
- Generation Phase: A POST request sends the JSON payload to Flask, which awaits the Groq API response.
D. Professional PDF Export (pdf_generator.py)
Logic: Converts the raw JSON from the AI into a structured document.
- Formatting: Uses
ParagraphStylesto apply academic formatting (Times New Roman style, bold question numbers, alignment). - MCQ Parsing: Automatically detects MCQ types to create indented [A, B, C, D] lists.
- Implementation: ReportLab's
SimpleDocTemplatebuilds the "story" sequentially, ensuring page breaks and headers are handled correctly.
✨ UI/UX Philosophy
- Glassmorphism: Using
backdrop-filterand semi-transparent layers to create depth. - Interactive Feedback: A dynamic Stepper updates states as you progress, and badges confirm successful file processing.
- Accessibility: High-contrast light text on dark backgrounds ensures readability for long documentation.
🛠️ Installation & Setup
- Clone & Install:
pip install -r requirements.txt - Environment:
Set
GROQ_API_KEYin a.envfile. - Launch:
python app.py
Developed with ❤️ as a premium tool for modern educators.