Researcher – University of Bridgeport / NASA

Wildfire Detection Using Vision Transformer

• Achieved 96.10% accuracy using ViT (vit_base_patch16_224)
• Processed and normalized a 10.74 GB image dataset for PyTorch
• Enabled real-time detection via deep learning + sensor systems
GitHub

Intern – IIT Guwahati

DeepSkinNet for Skin Disease Classification

• Built a 7-layer CNN achieving 99.8% accuracy
• Handled noisy and imbalanced medical data
GitHub

Intern – Silicon Valley 4U

Data Analyst

• Extraction of nexclap users data manually from Nexclap’s official website.
• The collected data is cleaned, organized, and transformed into a suitable format for analysis.
• Identified trends and patterns that could inform nexclap users’ projects, videos, and blog percentage.

Intern – Silicon Valley 4U

Python Developer

• Executed python programming with an embedded device.
• Prepared documentation and uploaded it to firebase cloud.
• Responsible for handling errors and updating code based on requirements

Teaching Assistant – DRK College of Engineering

• Guided student projects on ML topics: heart disease & fraud detection
• Assisted lectures, coordinated events, and supported grant submissions

Projects

Breast Cancer Detection Using RESNET-50 V2

• Conducted a study using the ResNet-50V2 deep learning model to detect breast cancer from the RSNA mammography dataset, focusing on enhancing early diagnosis accuracy.
• Demonstrated that ResNet-50V2 achieved high performance in classifying mammography images, indicating its effectiveness in supporting radiologists with accurate, timely assessments.
• Contributed to ongoing medical imaging research by showcasing how advanced CNN architectures like ResNet-50V2 can improve early breast cancer detection and patient outcomes.
GitHub

Quantum Kernel-Enhanced Hybrid Models for Start-Up Success Prediction

• Developed a hybrid Quantum-CNN model that integrates quantum kernel preprocessing into a classical CNN, using venture capital data from Crunchbase to classify start-ups as successful or unsuccessful.
• Demonstrated that Quantum Support Vector Machines (QSVMs) achieve competitive accuracy (~66%) with classical models (up to 68%), and the hybrid quantum-CNN model achieved 70% accuracy, outperforming standalone methods.
• Showcased the potential of quantum kernel feature mapping in enhancing learning on real-world, noisy datasets—paving the way for Quantum Machine Learning (QML) applications in financial analytics and VC decision-making.
GitHub

Heart Disease Prediction using Machine Learning Classification Algorithm

• Developed a heart disease prediction system using machine learning techniques such as Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN) to identify potential heart conditions at an early stage.
• Leveraged medical data mining and classification algorithms to uncover hidden patterns in patient records, enabling accurate prediction and clinical decision support based on historical health data.
• Validated the model on a Kaggle dataset with 1,225 instances, demonstrating improved precision and performance compared to existing methodologies, proving its effectiveness in real-world clinical scenarios.
GitHub

Identifying Fraudulent Transactions with Supervised Learning Models

• The model showed some trade-offs in precision for non-fraud cases, suggesting potential for improvement using advanced techniques like SMOTE or XGBoost.
• The model achieved an overall accuracy of 70%, with strong performance in detecting fraudulent transactions (Class 1).
• It obtained a recall of 0.79 and F1-score of 0.75 for fraud detection, indicating effective identification of most fraudulent cases.
GitHub