Dezhi “Andy” Fang

Enterprise Data Platform. Technical lead of trading-critical security master, providing timely and reliable access to fundamental reference datasets at Citadel.

• Architected and led the implementation of observability stack to enable consistent monitoring of service performance and health.
  • Integrated distributed tracing, unified metrics, dashboard, and alerting into several key systems. Set widely-adopted observability strategies.
• Designed and built organization-wide Kubernetes deployments tooling and best practices.
• Designed and led build-out of firm-wide, Spinnaker-based continuous deployment system for both Kubernetes and legacy NFS deployments.
• Modernized of build system for large-scale C++ (migration to Bazel) and Java (multi-project Gradle monorepo builds) services.
• Optimized performance and memory usage for large-scale C++ services, such as identifying memory fragmentation in glibc malloc(), which resulted in dramatic decrease in OOM and more than 75% memory usage reduction for a critical production service.

Cloud Infrastructure, Production Platform. Select projects include:

• Designed and led the build-out of Airbnb's distributed, delayed job scheduling and queueing system on top of AWS SQS and DynamoDB.
  • Read more about this work on Airbnb Engineering Blog: medium.com/p/93ab10f05f99.
  • Led an intern as intern manager during Summer 2019 for this work.
• Designed and implemented multi-cluster architecture for Airbnb's Kafka clusters, including a metadata proxy that multiplexes Kafka requests based on the Kafka TCP protocol.
• Implemented a thrift serializer/deserializer library for ruby that's more than 25x faster than the existing binding.
  • Read more about this work on Airbnb Engineering Blog: andf.us/sparsam. Find this work open-sourced at: github.com/airbnb/sparsam.
• Built distributed tracing pipeline for collecting frontend performance metrics.
• Built performance profiling and monitoring tooling for JVM applications running in Kuberenetes/Docker.

Research in the cross-section of data mining and human-computer interaction with professor Polo Chau.

• One of the investigators in NIH's MD2K (Mobile Data to Knowledge) initiative. Developed a predictive visualization dashboard for exploring correlations between activity, stress, and smoking relapses. Paper.
• MMap: Scaling up scientific computation with memory mapping. Paper. ARGO: Billion-scale visualization of network data. Paper.

Utilizing data mining (with Hadoop and Python) to improve malware detection efficacy.

• Inferring unknown files' reputations with parent file and execution environment features.
• Deployed new rule in Symantec's AV Engine with a false positive rate of only 0.2%. Used in Norton Security and Symantec Endpoint Protection.
• Implemented a decision tree executor for detecting malware using file features that replaced Symantec's existing implementation.