Projects

A growing collection of my projects.

research
Breast Cancer : Multi-Modal Fusion Dataset

Breast Cancer : Multi-Modal Fusion Dataset

A comprehensive data curation project addressing the fragmentation of the TCGA-BRCA program. This dataset provides a strictly patient-aligned fusion of radiology (MRI), computational pathology (WSI patches), RNA-seq, Copy Number Variations (CNV), somatic mutations, and longitudinal clinical records for breast cancer research.

PythonMulti-ModalMedical ImagingBioinformaticsTCGAKaggleWSIOmics
OncoGemma: Multi-Modal Oncology Agent

OncoGemma: Multi-Modal Oncology Agent

An autonomous virtual tumor board — fine-tuned MedGemma-4B with multi-modal fusion of MRI, whole slide images, and genomic profiles, featuring a dual-agent zero-hallucination inference engine for precision breast cancer treatment planning.

PythonMedGemmaLLMMulti-ModalLoRAOncologyFine-tuningNext.jsGoogleEdge-AIKnowledge-DistillationPyTorch
Medical Image Segmentation: Brain Tumor

Medical Image Segmentation: Brain Tumor

A systematic deep learning benchmark for automated brain tumor segmentation on BraTS20 MRI scans — comparing custom U-Net, pretrained encoder U-Nets, U-Net++, DeepLabV3+, and TransUNet across multiple loss functions and hyperparameter configurations.

PythonPyTorchMedical ImagingSegmentationUNetTransUNetBraTSComputer Vision
Edge AI Face Recognition System

Edge AI Face Recognition System

A production-grade, real-time face recognition system engineered for resource-constrained ARM devices — combining FaceNet embeddings, MTCNN detection, and direct Raspberry Pi GPIO integration for physical access control.

PythonPyTorchComputer VisionEdge AIRaspberry PiOpenCVFaceNetMTCNN
Medical Image Segmentation: GI Tract

Medical Image Segmentation: GI Tract

Deep learning pipeline for automated segmentation of gastrointestinal organs in MR-Linac scans, enabling precise radiation dose delivery for GI tract cancer treatment — benchmarking UNet, UNet++, and DeepLabV3 with multiple encoder backbones.

PythonPyTorchMedical ImagingSegmentationUNetComputer VisionKaggle
Lightweight Language Modeling: A Comparative Benchmark

Lightweight Language Modeling: A Comparative Benchmark

Comparative study of lightweight neural architectures for language modeling, benchmarking LSTM variants and Transformer-based models on WikiText-2 with a focus on perplexity and computational efficiency.

PythonPyTorchNLPTransformersLSTMLanguage ModelingWandB
Age Estimation in the Wild

Age Estimation in the Wild

Comparative study of three deep learning architectures for real-world age estimation, including MiVolo, ResNet, and Multi-Input GhostFaceNet — trained on IMDB, UTKFace, and CACD datasets.

PythonPyTorchComputer VisionDeep LearningAge EstimationWandB