Zero-Hallucination Assessment Engine
ExamCraft AI is a specialized assessment generation and split-screen crafting studio built specifically for secondary and intermediate education (Classes 9, 10, 11, and 12). Designed to eliminate AI hallucinations entirely, ExamCraft employs hybrid dense/sparse vector retrieval directly from verified textbook corpora.
Live System Diagnostics
Direct telemetry connection to the ExamCraft FastAPI backend
N/A
Disconnected
N/A
5-Stage Grounded RAG Pipeline
How ExamCraft converts raw textbooks into verifiable examination papers
PyMuPDF Text Ingestion & Parsing
Official Punjab Curriculum and Textbook Board (PCTB) PDF files are parsed using PyMuPDF. Text is chunked into logical units containing chapter headers, section titles, and exercise numbers.
FastEmbed Dense & Sparse Embeddings
Chunks are embedded using FastEmbed BAAI/bge-small-en-v1.5 (dense 384d vectors) paired with BM25 sparse vectors for hybrid semantic and keyword lexical retrieval.
Qdrant Vector Database
Stored in high-performance Qdrant collections. Hybrid search utilizes Reciprocal Rank Fusion (RRF) to rank the most syllabus-accurate text chunks for assessment generation.
Google Gemini 2.5 Flash Synthesis
Gemini LLM receives retrieved textbook chunks along with strict curriculum prompt constraints (Classes 9–12). It produces structured Pydantic schemas with MCQs, short questions, and long questions.
ReportLab Vector PDF Typesetting
The approved examination JSON is compiled directly into a publication-ready vector A4 PDF using ReportLab. Includes custom institution headers, student roll number blocks, 2-column MCQ tables, and full solution appendix.