Yichen Zhao

Ph.D

Computer Science

Dagshub Inc.

Data Science Intern

Next Word Prediction • Constructed a bidirectional LSTM network to predict subsequent words given text, harnessing MLflow’s model tracking capabilities. Reduced cosine embedding loss by 57.60% compared to baseline model by choosing the top-performing experiment from logged experiments. • Coached Dagshub users on practicing MLOps in their ML projects by publishing a blog post on the MLOps-integrated workflow of deploying the next-word-prediction APIs in the Dagshub community. Snowflake x Dagshub • Forged integration between Dagshub tracking and a Snowflake database by leveraging Git to log SQL queries and capitalizing on DVC to log result tables. Tuned the SQL queries to slash MSE of the Lasso regression model that forecasts Lego set prices by 69.32%. Documented the ETL workflow in a blog post to enlighten Dagshub users on utilizing Dagshub tracking features when collaborating with cloud-hosted databases on Dagshub. Annotation Transfer for YOLO-formatted Annotations • Automated the transition of YOLO-formatted annotations for rendering images with annotations in Dagshub repositories, slashing operation time by 80% versus open-source tools.