Experience

• 

  Vanderbilt university

  Aug 2017 - Jul 2022

  Research assistant

  1) Flight Anomaly Detection Using Hybrid Auto Encoder and One Class Neural Network (Python, Tensorflow, keras, Anomaly Detection) Fetched, cleaned, and smoothed heterogeneous time series flight data from NASA DASHlink Developed an auto encoder (AE) and one class neural network (OCNN) to detect the unlabeled anomalous flight data during landing phase Used Monte Carlo dropout out to estimate model uncertainty and obtain uncertainty bounds of reconstruction error Successfully detected the anomalous data points based on the proposed framework2) Landing Distance Prediction Based on LSTM Neural Network (Python, Tensorflow, keras, Time Series Forecasting) Used historical flight data to train a multi-step rolling prediction model to predict the landing distance for flights landed at Detroit Metro Wayne County Airport (KDTW) Developed multiple model evaluation metrics to evaluate model performance The proposed recurrent rolling prediction model significantly outperformed the state of art machine learning models3)Information Fusion and Bayesian Deep Learning for Hard Landing Assessment (Python, Keras, Time Series Forecasting ) Applied feature selection techniques on DASHLink data to reduce the workload of calculation which contains thousands of full trajectories and 186 features to 24 features Smoothed the data by forward and backward filtering with zero phase delay Developed Bayesian deep learning model to predict the vertical speed at touch down with additional model uncertainty information Optimized the proposed Bayesian deep learning model based on test data and expert feedback4)Data-Driven Modeling for Aviation Safety Diagnosis and Prognosis (Python, Matlab, Bluesky) Generated flight data by predefined command scripts using Bluesky simulator Built a systematic data-driven diagnosis and prognosis framework to detect anomalies in the flight trajectories and diagnose root cause Show less

• 

  Asurion

  Jul 2022 - now

  Data scientist

  Phone Claim Volume Forecasting (Supply Chain, Time Series Demand Forecasting) 1)Developed and fine-tuned 27 time series forecasting models, including moving average, ARIMA, SARIMA, and recurrent neural network models, using historical claim data 2)Selected model significantly enhanced forecast accuracy, leading to increased efficiency in the supply chain system and an anticipated total savings of $2MRed Flag Behavior Detection in Insurance Sales Text Modeling (NLP, LLM, AWS, Sagemaker) 1)Created and optimized six cost-effective Keywords-ChatGPT hybrid models to identify post-sales red flag behaviors such as service exaggeration and pricing discrepancies 2)Tailored each model to target specific types of red flag behaviors 3)Achieved model precisions ranging from 65% to 75%, surpassing baseline benchmarks set at 50% 4)Established the model deployment pipeline and automated daily job execution through Amazon EventBridge Show less

• 

  Vanderbilt university

  Sept 2023 - now

  Adjunct assistant professor

  implemented impactful teaching strategies, incorporating regular recap sessions, utilizing Top Hat for real-time assessment and participation points, and making timely adjustments to course materials based on student performance. Addressed coding challenges for data science students by breaking down code blocks, fostering independent problem-solving, and promoting inclusivity through personalized teaching and support for diverse backgrounds and skill levels