Databricks & LLM Data Engineering

Databricks

Hadoop

CHI WENG (MICHAEL) CHEONG

516-***-**** Suffolk, NY ********************@*****.*** Please checkout my LinkedIn profile: https://www.linkedin.com/in/chi-weng-michael-cheong-60b47350 Thorough understanding of neural networks and LLM. Strong database and python programming skills (web app backend development and machine learning) and rich hands-on coding experience. Expertise in data transformation, data analysis and prediction with data. A solid foundation in computer science, such as algorithm and data structure. Over 6 years research experience in quantitative science including mathematics. SKILLS

• Deep learning: CNN, Transformer

• Programming skills: Python, Pytorch, Pyspark, FastAPI, GraphQL(strawberry), Flask, Cypher (Neo4J), SQL, Mapreduce, Spark, Java, R* (* stands for coursework)

• Analytic tools: Pandas, NLTK, Scikit Learn, Weka, R*, Spark, Scala

• Devops tools: Jules (based on Jenkins)

• Techniques: Natural Language Processing, Graph Analytics, Classification, Clustering, Regressions

• Platforms: Neptune (based on S3), Talend, Linux, Git, Athena (based on CVS), Azure Cloud Platform, Azure Platform, Athena (JPMC proprietory), Gaia (JPMC propriety platform)

• Cloud Platforms: Azure SQL, Azure Streaming Analytics, Azure HDInsight Cluster, Azure Web service and Azure Databricks (spark)

• Citizenship: US citizen

EDUCATION

M.S. in Predictive Analytics and Data Science, St. John’s University, NY May 2020 Ph.D. in Mathematics, Purdue University, IN June 2013 B.S. and M.Phil. in Mathematics, The University of Hong Kong June 2007 EXPERIENCES

Independent Researcher on Small/Large Language Models

• Assess website based LLM memory system

• Experiment GRPO on Tencent llama pro type models to achieve performance gain from the newly added layers. Senior Software Engineer PGIM Fixed Income Apr 2023~June 2026

• There is the challenge of messy SQL statements everywhere in the company and no place to centralize them. Build a data access layer for traders to use instead of complex SQL queries. It stores the select statements of queries in a database and make them intercallable. Users by selecting the select statement they want get data from database as from statement is automatically generated. It was python backend using fastapi and graphQL. It also has heavy use of lark and OOP in python.

• There is challenge of manual work in gathering information about bond issuance information in various countries. Web scraping of bond issuance data to consolidate bond issuance data from various country with various sources

(png, pdf, html, xls) into a single table. Need to keep the data accurate and up to date. It’s a python project with heavy use of pandas and multithreading. It also has heavy use of pdf and png parsing. Created a mechanism for AI to read data that should generalize to massive data but so far only have 100% accuracy on 12 pages pdfs and 3 jpeg files.

• There is the challenge of lack of performance for very complex SQL queries written by quants. Adhoc requests for SQL query optimization. It has heavy use of various SQL optimization techniques on very complex queries. Also optimized complex stored procedures.

• Rewrite Quant model preprocessing ETL pipeline from Pentaho to python

• Create regression testing framework for testing new quant models

• Maintained a 140k+ lines C++ code in three week without documentation

• Was working on converting autosys perl scripts to python Software Engineer Google Dec 2021~April 2023

• Using Python and JavaScript to develop observability tools and writing script from Git repos to deliver conflict- over-time statistics to the database. Additionally, conducting data visualization for insights using Google proprietary visualization tools.

• Developed Chrome browser extension that captures metrics and sends information back to Google Analytics. Tool provides performance and page load-time metrics. It was JavaScript for the frontend and Java for the data processing.

Senior Software Engineer Bloomberg Jan 2021~Dec 2021

• Migration of metering to mapreduce: migration of metering of billions of customer records from Python+SQL setting to Java+Mapreduce setting. The code base is around 30000 lines of code and worked on for multiple years by more than 10 developers.

• Devops : Set up Jenkins pipeline for code deployment on a spark project of comparing metering results. Software Engineer JP Morgan Chase Nov 2018~Jan 2021

• Machine Learning and WebApplication: 1) Leading the web application development team (with 2 people) for 3 months

(as technical team lead) then worked as a developer, on a project providing automatic email classification by nlp machine learning techniques. Duties include maintaining data science projects, work on infrastructure, production release and adhoc data requests. Worked on fixing machine learning defects mainly on entity extraction based on regular expressions. The project was done in Python.

• Devops : Utilizing internally developed product Jules based on Jenkins to set up automatic code scan and deployment. Migrated from aim or Athena to Jules based deployments on various projects.

• Infrastructure : 1) Investigating the connection between Jupyter as a service (based on docker technology) and an on premise Spark Cluster. Established (partially) and tested access to hdfs, hive and yarn from Jupyter as a service. Loading manual classification data for data scientist use and automatic reporting. 2) Worked on improving Athena platform’s support for Flask to enable healthcheck, load balancing from restarting services. Also, worked on various infrastructure tasks like mounting NAS share, ordering of load balancer, setting up Jules.

Principal Platform Engineer CA Technologies July 2017~Nov 2018

• Machine learning part of Data Pipeline : This project retrieve and push data to and from Azure SQL and Azure Datawarehouse. Part of the ingestion involves Azure Streaming Analytics.

• Web Application: 1) worked as backend developer of Campaign Planning Tool project to provide a unified sales and marketing platform for the whole company. The project is written in Python with flask framework using Neo4J. This project runs on Azure Instances. 2) Maintaining (bug fixing and adding new functionalities) the backend for a Contact Intelligence project that provides web service UI for user to navigate through sales contact data.

• Test/Evaluation: Developed testing tools for Contact Intelligence in python by replacing Neo4J with Pandas. Evaluated graph algorithm performance between Neo4J and Cosmos DB graphs. OTHER ACCOMPLISHMENTS

Cheong, C. W. (2015). Unitary orbits in 1-dimensional NCCW complexes. Journal of Functional Analysis, 269 (9), 2977-2994. PROFESSIONAL DEVELOPMENT

Deep learning course and ethics in AI course at UT Austin

Contact this candidate