Mohammed Abdul Al Arafat Tanzin

A production-grade Retrieval-Augmented Generation (RAG) system that transforms a personal portfolio into an interactive AI assistant. Implements document processing with recursive text chunking, Gemini embeddings (768-dimensional vectors), and Qdrant vector database for semantic search. Features a modern React frontend with TypeScript, FastAPI backend, and multi-model fallback for reliable response generation.

Developed an automated video anonymization pipeline using Grounding DINO and OpenCV to detect and blur vehicles, license plates, human faces, and heads in Kuala Lumpur road scenes. Implemented GPU-accelerated batch inference and non-maximum suppression for efficient video processing.

Advanced multimodal sentiment-based text summarizer that analyzes emotional context to generate more meaningful summaries.

Developed a hybrid ML framework combining sentiment analysis of user comments with engagement metrics to mitigate misleading like/dislike ratios.

Designed ReCAN: an efficient CNN-attention hybrid using ResNet-18 base + alternating skip connections and channel attention. Achieves 8–12% higher accuracy while remaining lightweight (~14.47M params).