* Developed and Deployed Python packages and Docker images to productionize machine learning models which support real-time product tagging with a speed of 1K+ requests/second.* Leveraged Apache Spark with PySpark to parallelize data processing for 100M+ competitor products to produce competitor product tagging results.* Served as a tech lead to manage a team of three to tackle product tagging: ** Applied transfer learning to train a word embedding model for product centric corpus and identify 20K+ synonyms for product tagging keywords using nearest neighbors. ** Leveraged Market Basket Analysis to identify correlation between product tags and build product tagging graph.

* Leveraged fastText to build and deploy a two-stage product classifier system improving the overall classification performance of product classifier by 30%.* Utilized HDBSCAN clustering to restructure the product tag taxonomy used by 4.5M+ products. * Developed and deployed machine learning (ML) and non-ML models to achieve automated product tagging with various approaches tailored to the corresponding problem. Achieved precision score of 78% - 96% : ** Aho-corasick string matching algorithm driven information extraction. ** Natural Language Processing powered text based classification. ** Convolutional Neural Network powered image based classification. ** Ensemble models to perform classification with product descriptive data and imagery data. ** Conditional Random Field* Created a Python package to visualize and explain the prediction of Keras-based image classifier.* Built automatic pipelines using Airflow to retrain and deploy machine learning models with cadences.

• Built http://drbrain.online, a tool for accelerating neurodegenerative disease diagnosis with 3D brain MRI images.• Utilized Python and Linux command line tools to process 140 GB of image data. Cleaned up 3D brain images by fixing their orientation, cropping and segmenting them, and applying affine transformation.• Set up AWS EC2 instance (g2.2xlarge) for deep learning implementation with Keras, Tensorflow and GPU backend.• Created an anomaly detection system with Isolation Forest algorithm to increase user experience by detecting and filtering out invalid MRI images.• Developed, trained and validated a 3D convolutional neural network model with inception modules for classifying brain images as healthy, Alzheimer's disease, or Parkinson’s disease. Achieved accuracy levels of 76.8%.

• Implemented metabolomics and next generation sequencing techniques (ChIP-seq and RNA-seq) to elucidate molecular mechanisms of metabolic side effects induced by steroid drugs. This work resulted in five publications and numerous conference presentations. • Engineered a Chromatin immunoprecipitation-sequencing (ChIP-seq) analysis pipeline system, TBMI_ChIP (https://github.com/tcchenbtx/TBMI_ChIP), to consistently automate ChIP-seq analysis.• Discovered three novel molecular pathways to reduce hepatic insulin resistance and lipid metabolism disruption induced by steroid drugs.• Conducted a reproducibility analysis on a functional magnetic resonance imaging (fMRI) based image-processing research study.• Applied linear modeling and time series analysis to study brain fMRI data for visual pattern recognition. (https://github.com/berkeley-stat159/project-zeta)• Mentored 45 undergraduate students in Computational Toxicology, guiding them to perform data mining on various websites and databases to discover novel insights in the field of toxicology. Several students went on to present in the National Society of Toxicology Annual Meeting.