Developed an LSTM-based time-series model for Tesla stock price prediction, reducing testing RMSE by 30% compared to Logistic Regression and XGBoost. Conducted EDA, feature engineering, and created interactive dashboards with visualizations (line charts, heatmaps, AUC-ROC) to improve model.

Developed a sentiment analysis system using deep learning, achieving 72.81% adjusted R² and 80% accuracy in poverty analysis. Used Pandas and R for preprocessing, modeling, and validation. Enhanced precision through probability threshold optimization.

Built and evaluated ML models (KNN, Logistic Regression, SVM, Decision Tree) on a credit card fraud dataset with 284,808 transactions. Used PCA for dimensionality reduction, SMOTE for class imbalance, and cross-validation for tuning. KNN and Decision Tree achieved over 90% accuracy.

Built and fine-tuned CNN and ResNet-18 models for bone fracture detection from X-rays, achieving 80.1% accuracy using transfer learning. Improved performance by addressing underfitting, overfitting, and class imbalance, showcasing AI’s impact on medical imaging.

Built a CNN model with TensorFlow to detect distracted driving in real-time, trained on 22,424 images. Optimized for in-car cameras and enhanced accuracy using Voxel51’s FiftyOne and image preprocessing.

Created 'RecruitMatic,' an AI-powered platform for recruiters using OpenAI API, Python, and Streamlit, boosting performance by 30% and document parsing by 35%. Designed an intuitive UI to streamline workflows, enhance candidate insights, and improve data extraction accuracy, reducing response times.