• Built a feral hog detection model to help North Texas farmers to prevent appr. $2.5 B crop damage every year.
• Developed an LSTM transformer dart model using PyTorch for multi-factor time series forecasting.
• Automated end-to-end Machine Learning model deployment with CI/CD integration.
• Designed video processing models for drone detection and performed human motion tracking using computer vision tools.
• Worked under the Industrial & systems Engineering department to develop repositories and manage data for research.
• Built a Pipeline for end-to-end data transfer from snowflake to big query using GCP cloud function with pre-processing, feature engineering/selection.
• Performed Clustering and statistical analysis to find relationship between factors leading to risky sexual behavior risks using Nursing data.
Programming : Python, HTML, CSS, Java, C, C, MATLAB
Data Engineering : SQL, Snowflake, HQL, Presto, Db2, MongoDB
Tools : Git, GitHub, Visual Studio, Power BI, Tableau, ERP, Minitab, YOLO, OpenCV
Big data & Cloud : GCP (Airflow, Cloud Functions, Cloud Build), Azure (AutoML), Spark
Unmanned Aerial Vehicle (UAV) Detection System - The US Air Force, #YOLOv5 #Object Detection
• Reduced manual inspection time and helped enhance the security capabilities of the US Air Force by building a real-time drone detection and monitoring system (funded project).
Real-time Speech Emotion Recognition for Autism children, #Social media data #NLP #HPC
• Built a ResNet-based neural network for audio classification and to predict emotions.
• HPC clusters were utilized for parallelized model training.
Lending Club – Loan Default Prediction, #Predictive modeling #Machine Learning
• Extracted, Pre-processed, and manipulated raw data from Lending club. Built Predictive Models such as Logistic Elastic net, Random Forest & SVM. A user-friendly Power BI dashboard was created to check Loan status.
Netflix - A Movie Recommendation System, #Machine Learning #Recommendation engine
• Implemented a hybrid recommendation system using Collaborative filtering and Pearson’s R correlation for suggesting movies for users based on prior user ratings. Achieved recommendations identifying pairs using Jaccard similarity.
Credit Card Fraud Detection, #Anomaly Detection #Neural network
• Used SMOTE to overcome class imbalance and PCA & SVD to reduce feature dimensions. Trained and tested using Random Forest, XGBT, Neural network models with 98% accuracy.
Real-time Golf Pose Correction, #Human Motion #Pose estimation
• Combined MediaPipe and ZED cameras for precise real-time golf pose correction, capturing crucial body data to boost performance through immediate feedback on posture and swing mechanics.
Amazon’s Data Structures certification || Google Data Analytics certification || Computer Vision and Image tools certification.