• LLM agent development , multi-turn SFT and agentic RL training, evaluation

• Python backend service performance improvement

• Maintain multiple data pipelines with pyspark

• Designed and Developed an open-source, semi-automated data ingestion tool based on excel spreadsheet and CRIPT sdk for chemists to upload experiment data to CRIPT database using python. Designed the tool with a compiler-like structure. Assisted early users to upload about 100 thousand data points to database.

• Created unit tests for the data ingestion tool, fixed 20+ bugs based on self testing and user feedback

• Developed ETL pipeline with 30+ dependencies to process collected raw data and store in Hbase using HiveSQL. Developed 4 DM tables with different aggregation degrees to satisfy needs. Separated DW data from business logic. Supported the data needs of 200+ colleagues from data science team and operation team.

• Maintained event tracking ETL pipeline and report 10 bugs to front-end developers to ensure data quality.

• Improved the query efficiency about 50% by implementing database sharding strategy on a 2000Gb database.

• Collaborated with team members to implement a regression model to evaluate GMV increment of sales promotion tools using SQL and python and visualized results with matplotlib. Deployed the model to perform daily update.

• Developed timing tools to automate calculation of business indicators and discern abnormal changes using python.

• Analyzed the user’s behavior in each step of registration process using PrestoSQL. Feedback was given to the product manager and updates in the registration process was carried out, which reduced the CAC by 10%.