At ARDENT Computech, I developed an end-to-end deep-learning pipeline to classify brain tumors from MRI scans. Using a pretrained ResNet101 backbone, TensorFlow, Keras, and OpenCV, the model achieved 92.4% validation accuracy across four tumor categories.
I also developed image preprocessing and augmentation workflows, applied transfer learning and regularization to reduce overfitting, and optimized training speed by 28% while preserving classification performance. I implemented confidence-aware inference and assessed predictions against labeled validation data.
For InkTrace, I built a whiteboard-image dataset with controlled erasure variants and pixel-level residual-writing masks. I implemented and evaluated a Residual U-Net, then developed a predicted-mask-guided enhancement pipeline and integrated EasyOCR for text extraction.
I also designed an NLP workflow for context-aware synonym substitution that aims to preserve sentence meaning. My publication, “AWLR: Attention Weighted Logistic Regression for Fake News Detection using Named Entity Recognition and Semantic Similarity Measures,” lists me as a co-author.

