Hi, I am Vincent Huang, a software engineer at Databricks.
I graduated from McGill University with a Bachelor's degree in Honours Computer Science and a minor in Musical Science & Technology. And I hold a Master's degree from Carnegie Mellon University in Machine Learning and Natural Language Processing. My career started in telemetry infrastructure, where I worked on the nexxt-generation log delivery system at Databricks. I recently transitioned to focus on observability products, where I adapted my prior experience to build industry-leading query performance insights that scale with second-level latency.
My interests lie in observability products and infrastructure in large-scale distributed systems. And I am also an avid classical guitar hobbyist.
Databricks
- Working on industry-leading products that improve Spark query observability and surface user-facing performance insights.
- Built the next-generation log delivery infrastructure from the ground up, which increased the data completeness from 99.9% to 99.999%.
Apple
I joined as the first machine learning engineer on the team.
Major highlights of this internship:
- Service Facade: Lead the end-to-end development of a proxy layer with Spring WebFlux from scratch used by 4 teams while providing advanced customizable features (Rate Limiter, Circuit Breaker) for each of the upstream service.
- VoIP QoE: Research, design, and integrate Quality of Experience(QoE) metrics for call sessions with Machine Learning into Apple’s Retail Platform.
- Deployment and Monitoring: Integrate the application into Kubernetes cluster, set up ACL and security groups. Set up Splunk Stats Services with application.
Unity Technologies
I worked with the Deeppose team in the Unity Labs to help address the pose estimation problem from inverse kinematics using the ProtoRes Network.
Major highlights of this internship:
- Knowledge Distillation and Model Architecture: Designed model architecture which, combined with knowledge distillation, achieving comparable performance to the original model proposed by Harvey et al with only 1/20 of size
- FPS Improvements: Supported 5 times as many frames a second as the original model by improving model inference time
- ML Engineering: Adapted Knowledge Distillation algorithm to motion synthesis tasks
Unity Technologies
I worked with a team of 5 to build and improve the Unity Package Manager
Major highlights of this internship:
- Provided user-facing, event-oriented download progression report for Unity packages
- Unified installing and caching pipeline for packages from different sources
- Established decentralized testing framework for the whole team while maintaining 100% code coverage
- Improved search reply query, which cut the client-side package search time