• Designed event-driven toolkit for non-standard data such as financial text data. Mined factors and designed strategies via event-driven methods• Developed factor pipeline to generate a batch of factors from non-standard text data based on ML and DL methods

• Mined daily frequency factors. Factors had significant Sharpe and annual return, with low correlation and were put into production• Added analysis modules into original backtesting framework, including risk attribution module and factor distribution module etc.

• Developed NLP model based on Longformer and conducted incremental training to better extract information and predict return• Conducted distributed training to solve inefficiency of long text training task. Reduce training time to one-third