Commercial Applications

  • I built an automated sales pipeline using machine learning for a client. Estimated customer return dates for auto dealers all across China.
  • Deep learning in industrial defect detection. I researched, created data collection pipeline (industrial cameras and lighting setup), collected and labelled data, trained a CNN and delivered a proof of concept to a client. System performance was on par with human quality assurance employees.

Deep Learning

  1. Eye tracking system. I built a web app for data collection and then trained a deep learning model that predicts which part of laptop screen a user is looking at. Together with a group of friends we productized this technology at a Beijing Startup Weekend in May 2019 and got the third place. [Python, fastai, Flask]
  2. Leafy greens classifier. I often feel confused when shopping for leafy greens in China. This is why I duilt a classifier to tell them apart. You can try the deployed model here. [Python, fastai, Docker]


  1. Stock portfolio optimization using SVM. I developed a novel method for selecting and optimizing low-risk stock portfolios using support vector machines. The model outperformed SP500 benchmark by 15% in terms of return per unit of risk [Python, scikit-learn, CVXOPT]
  2. Sentiment analysis of Twitter feed to predict stock market movements. I analyzed 200 gigabytes of Twitter feed data to extract sentiments of individual tweets. Then I built a sentiment index that was predictive of stock market movements [Python, R]

Autonomous Driving

  1. Lane detection using computer vision techniques [Python, OpenCV]
  2. Traffic sign classification using deep learning [Python, Tensorflow, OpenCV]
  3. Behavioral cloning using deep learning [Python, Keras, OpenCV]
  4. Advanced lane detection with camera calibration. In addition road curvature and vehicle offset relative to the center of line are calculated [Python, OpenCV]
  5. Computer vision-based real-time vehicle detection system [Python, scikit-learn, OpenCV]


  1. Predictive modeling in breast cancer diagnostics using supervised machine learning techniques [Python, scikit-learn]

Deep Learning: Non-Technical Writing

  1. CapsNet Series in four parts (1, 2, 3, 4) — a popular introduction of the novel neural network architecture. This series was originally published on Medium (1, 2, 3, 4), where it got a lot of attention.
  2. Stochastic Weight Averaging — a new way to get state of the art results in deep learning.
  3. Deep Learning Vision for Non-Vision Tasks — three case studies about creative application of deep learning vision models to non-vision tasks
  4. Key Deep Learning Architectures for Visual Object Recognition — a series of posts about the most important CNN architectures.