United States

Back-end

Data engineering

AI/ML Engineering

Architectural Engineering

Carnegie Mellon University

Wuhan University

Bowen Deng

Software Engineer

Problem-solving

AI/ML

Cloud computing

Machine learning

Integration Testing

Unit Testing

Pandas

Python

PyTorch

Ruby

Airflow

Apache Spark

D3.js

Docker

GitHub

HTML

Java

JavaScript

Jira

Jupyter Notebook

Keras

Matplotlib

Numpy

TensorFlow

REST API

MongoDB

Redis

Node.js

Kafka

• Design, implement and operate the internal infrastructure for machine learning.

• Led the architecture and implementation of an in-house machine learning platform that streamlined the entire end-to-end process, including data ingestion and pre-processing, model training and optimization, performance evaluation, and continuous model deployment.
• Developed and implemented image similarity models for retail items using semi-supervised learning techniques to overcome noisy labels, resulting in a 12% improvement in F1 score compared to competitor's classification approach.
• Optimized input pipelines for large datasets by leveraging Bigtable, achieving a 2x speedup for model training.

• Designed and implemented rstream, a stream processing library that supports automatic state persistence and failure recovery with reference counting
• Enhanced in-house tracing infrastructure around gRPC and Kafka, making profiling and debugging more accessible across services
• Developed event-sourced microservices for data ingestion and aggregation
• Developed and maintained CI/CD pipelines on Azure DevOps for production services
• Implemented libraries and development tools to support remote debugging for production services (written in Node.js or Python) on Kubernetes, and significantly shorten troubleshooting time for several production issues

• Built a system detecting landing page cloaking (showing reviewers benign pages, but redirecting users to other malicious pages), and the system enforced 130K+ Ads within the first week of its release with a 0.04% false positives rate 
• Developed a time-windowed counter service processing 12K log events sampled from mobile clients in real-time and aggregated the events to high-level features for cloaking detection

• Developed Cosmos pipelines analyzing large scale (TB-level) semi-structured transactional logs from Microsoft Azure and Office 365, and helped the product teams caught several system failures in their early stage
• Analyzed performance characteristics of distributed jobs for log analysis, identified performance bottlenecks caused by poor data locality and redundant computation, and performed optimization archiving 10x speedup
• Designed an extendable log analysis framework and integrated several existing log-analysis algorithms into the framework

General Mentorship Session

Bowen Deng🇺🇸

About Bowen Deng

Expertise

Sessions with Bowen Deng