Projects

A Machine Learning RAG (Retrieval-Augmented Generation) AI Chatbot designed to assist with research. It leverages advanced NLP to query documents and provide accurate, context-aware answers via a Gradio interface.

Developed deep learning models (YOLO, ViT, MobileNetV3) to classify casting products as defective or non-defective with 99.9% accuracy. This project involved collecting real industrial data, annotating defects, and training robust models for automated quality control.

Designed and fabricated a precision tailstock die tool holder using SolidWorks and lathe machining. The project focused on ensuring high dimensional accuracy for threading operations on a lathe machine.

MobileNetV2-based classification system for 12 local fish species, deployed via a Streamlit web application. Aimed at assisting local fishermen and researchers in rapid species identification.

Pixel-level segmentation of Glioma, Meningioma, and Pituitary tumors from MRI scans using YOLO architectures. The system assists radiologists by highlighting tumor regions with high precision.

Real-time compliance detection system identifying PPE and safety goggles. Designed to monitor factory floors and ensure workers adhere to safety protocols to prevent accidents.

Automated background removal tool using U²-Net (ONNX) with a Streamlit interface. Allows users to upload images, remove backgrounds instantly, and replace them with custom colors or images.

Predictive modeling of machining surface roughness using various ML algorithms. The Decision Tree model achieved an R² of 0.85, helping optimize cutting parameters for better finish.

Forecasting wind speed and direction in Dhaka using ANN, RNN, LSTM, and CatBoost. Provide data-driven insights for wind energy feasibility studies in the region.

Real-time live tracking of sign language gestures using webcam input. Utilizes computer vision techniques to interpret hand movements/gestures for automated recognition.