Manish Kumar Roy

Developed and deployed an industrial-grade real-time nut detection system leveraging 2,254 images, achieving 99.5% mAP@0.5, 99.8% precision, and 99.5% recall with YOLOv8s.

Implemented robust data cleaning, brightness normalization, CLAHE, data augmentation, and adaptive confidence thresholding to ensure high model performance and reliability.

Reduced manpower by 20%, enhanced production accuracy by 17%, and accelerated inspection speed by 99.4% through strategic system deployment.

Built a personalized news recommendation system using collaborative and content-based filtering, leveraging Twitter activity and reading history for relevant article suggestions.

Applied transformer-based BERT embeddings for sentiment analysis and K-Means clustering to optimize recommendations for over 10,000 user interactions.

Increased user engagement by 20% and boosted click-through rates by 15% through data-driven recommendation system improvements.

Collected and preprocessed 30k data points on linear, circular, and rectangular laser welds, conducting real-time experiments for predictive modeling.

Implemented and optimized Polynomial Regression, Decision Tree, Random Forest, XGBoost, and LSTM models with optimized hyperparameters.

Achieved an R2 score of 0.98 with Random Forest for temperature prediction and 0.85 with LSTM for load data.

Enhanced overall battery joint performance by 10%, reducing EV failure risk through real-time weld strength analysis.

Analyzed a comprehensive dataset of 20,500+ images across 15 plant species, labeled with various diseases and health statuses.

Applied data augmentation techniques including resizing, cropping, and normalization to enhance deep learning model robustness.

Utilized deep learning architectures like ResNet50, VGG16, VGG19, and InceptionV3, achieving 98.5% accuracy with VGG19.

Achieved 97.5% precision and 98.1% recall on the test dataset with VGG16, outperforming traditional methods.