Key Qualifications & Responsibilities:Driving the interaction between managers to ensure the analysis needs and generate the pull-through of insights of business.Data pipeline: building pipeline for data cleaning, including deduplication, treating missing data, and getting statistical summary, join across multiple complex data sets using SQL and Python. the efficiency was lifted at least 50%. Feature selection: Fixed overfitting problem by using coeff and feature_importances_. Selecting 6 features from 2100 features to represent 82% importances in a large project.Developing Automating ML model building using PyCaret and MLflow to reduce the debug time with 85% and predicting stomach disease classification for a large provider.Working in an agile environment within Google Vertex AI cloud. Collecting products historical data, metadata and reviews / comment data from Amazon using RainForest API and Querying BigQuery cloud database using SQL.Driving the interaction between managers to ensure the analysis needs and generate the pull-through of insights of business. Developing a NLP semantic search model to search unstructured text databases. Increasing the search speed by more than 10 times. Creating visualization graphs, like scatter, bar graph, heatmap for demonstration purposes, present results to professional and non professional people.

Developed machine learning and deep learning algorithms on healthcare cancer image classification and segmentation, developed NLP/NLU algorithms for electrical text records. NLP. Classified electronic health records to predict cancer using random forest methods. Sklearn, Fastai for torchtext were used as tools. Pre-trained BERT model gave me 89% accuracy. Skin cancer detection. image transfer learning using resnet / mobilenet-2 as backbone, Unet and Mask R-CNN as models, within Torchvision, Pytorch, Tensorflow, Fastai and Detectron2. Predicted skin or lung cancer, detected the localization. Visualization. created geographic interactive plots using plotly, folium, seaborn and matplotlib, presented confusion matrix heatmap, top-k-losses on multiple projects, including cancer image and text, COVID19 dataset.

Developed and Automation models: using Gradient boosting and Random Forest within Sklearn and modern deep learning frameworks, Keras, TensorFlow and Pytorch. Data Cleaning: data cleaning, transformation, preprocessed (NLTK) before modeling. Sentiment analysis of NLP: Implemented GloVe, TF-IDF, Word2Vec, pre-trained-BERT Models for projects. Modeling: RNN, LSTM. Random Forest, Gradient Boosting. Tuning hyper-parameters: n_estimators, max_depth, min_samples_split, min_samples_leaf, max_leaf_nodes, learning rate, dropout regularization SAS: Managed and delivered multiple productions, improved SAS SQL programs to more efficiency.

Developed SAS macro, SQL and Python predictive modeling on large datasets. Worked closely with development teams to ensure accurate integration of machine learning models, mentored junior people and helped them creative thinking in the work. Code optimization and Program development: Leaded the production development. Both STRATS and CCIP projects were optimized to reduce process time from 8-9 hours to 15 min, 6-7 hours to 1 hour respectively. ETL: Extracted data from Oracle or SQL server using Toad or SAS pass through. Did extensive data manipulation and transformation, batch loaded files to database. Predicted default, fraud, delinquency and foreclosure using Tensorflow The goal of this analysis is to predict if the probability status moved from DPD30 to DPD180 status. Dataset contained delinquency status such as DPD30, DPD60, DPD120, DPD180,... , explored data and features selection (RFE / RFECV), RF is the best model among multiple models for predicting the probability status changes among the statuses.

Outlier Detection on medical EP data to predict autism disease.

Manipulated data using Igor Pro / Matlab /SQL / SAS. Conducted ad-hoc and post hoc on medical data and predicted the open probability.