Responsibilities include:

• Internal Document Understanding Multi-Modal Framework with HuggingFace Ecosystem
• API Documentation
• Guides for Data Loading, Pretraining, Finetuning, etc.
• Explored AutoML’s Potential in Text/Vision Transformer Models
• Content-based Recommendation Feature for Existing Document Recommendation System in Adobe Acrobat

Responsibilities include:

• Neural Architecture Search Space
• Supervised/Unsupervised/Semi-supervised Detection Algorithms
• BlockChain Transaction Analysis
• Payment Fraud Detection
• Cyber Security Intrusion Detection

Responsibilities include:

• Action Detection System for elderly care facilities
• Safety Gear Detection System for construction companies

• An end-to-end system that supports easy pipeline construction with more than 70 primitives for automated machine learning.
• Top 3 contributor.
• Mentor for new team members;
• Explored neural architecture pipelining combination.
• Open sourced on GitHub, with 450+ stars and 50+ forks.

• Deep learning project for action and emotion detection used in ”Smart Homes”.
• Detected actions including coughing, hand washing, falling, cleaning windows, cleaning bathroom and washing feet.
• Implemented with Keras, used Kinetics 700 dataset, built VGG16 and Xception CNNs for base model.
• Found in total of 23000+ clips from 800+ YouTube videos, average accuracy 91.2%, ranked top three overall in the project competition.

• Data analysis and visualization on COVID-19 dataset from CSSE at John Hopkins University.
• Implemented with Scikit-Learn, Plotly, TODS and Tensorflow.
• Utilized algorithms including K-Means, DeepLog(Outlier Detection).

• Re-implemented the Neural Cleanse baseline system with Tensorflow.
• Explored several types of masked trigger including local and global features.
• Found global masked triggers to successfully inject models while not getting detected.

• Detection system to generate personal playlist based on age and mood.
• Used Flickr & YouTube Music APIs.

• Web, iOS and Andriod application for tracking & delivery of fuel trucks, deployed on Google Play Store.
• Implemented using Google Maps API and OpenWeather API, programmed in Dart language.

Constructing machine learning pipelines for outlier detection in time series with automated machine learning.