Project Intern at Larsen & Toubro Limited

Engineered automated PDF table extraction pipeline using custom regex patterns, Python, and NLP techniques, processing 500+ documents and improving data parsing accuracy by 20% for manufacturing workflows. Built intelligent keyword detection system with targeted pattern matching algorithms, reducing manual verification by 35% and accelerating document processing speed by 3x. Collaborated with cross-functional teams of 15+ members to integrate automation tools with existing quality management systems, delivering customer-facing features ahead of schedule. Optimized data extraction workflow with Pandas and NumPy, enabling seamless integration with production databases and reducing data entry errors by 45%.

Research Intern at Vivekanand Education Society’s Institute of Technology

Curated and preprocessed domain-specific dataset of 5000+ labeled samples using Python, OpenCV, and data augmentation techniques, enhancing model training reliability and achieving 91% augmentation rate. Implemented and benchmarked 3 state-of-the-art architectures (CNN+LSTM, I3D, YOLOv8+ResNet) using TensorFlow and PyTorch, optimizing for accuracy, precision, recall, and F1-score across 10 performance metrics. Achieved 93.87% accuracy, 0.9376 F1-score, and 94.2% precision with YOLOv8+ResNet hybrid model, outperforming baseline models by 12% and demonstrating production-ready performance for real-time deployment. Optimized model inference pipeline through quantization and pruning techniques, reducing prediction latency by 40% (from 150ms to 90ms) while maintaining high accuracy for real-time applications. Conducted comprehensive model evaluation with confusion matrices, ROC curves, and cross-validation, generating detailed technical documentation for research publication.