• Hybrid Model Registry

To enable a seamless hybrid cloud experience by integrating model registries across different cloud environments, ensuring smooth model management and deployment across different cloud environments.

• Model Registry Public-Facing API

Led the design and implementation of Cloudera's AI Registry public-facing API, a critical foundation for MLOps in the serving platform. This registry integrates with MLflow and allows importing external models, including Hugging Face and Nvidia NIM models, to streamline model management and deployment.

• Business User Experience for Cloudera Machine Learning

Delivered a seamless, intuitive interface for business users, improving accessibility and driving adoption of Cloudera Machine Learning solutions

• Conducted Proof of Concept for Feature Store Integration

Led a Proof of Concept (PoC) for integrating Feast, an open-source feature store, with Cloudera Machine Learning. As part of this effort, evaluated various databases for online and offline stores, including PhoenixDB, HBase, Impala, and Iceberg, ensuring optimal performance and scalability for feature engineering workflows.

Lifecycle management server for Cloudera Data Science Workbench (CDSW) running on public cloud (Cloudera Data Platform).

• Provision cloud resources (i.e. EFS) and install CDSW to customer's cloud environment

• Suspend/Resume CDSW with cloud autoscaler

• Upgrade Kubernetes for CDSW

POC for Istio service mesh integration with CDSW

• Automated Istio / Istio CNI installation

• ~90% mutual TLS coverage

Cloudera Data Science Workbench (CDSW) - A web application composed of Microservices written in go / node.js running in a K8S cluster.

Worked on general backend development for on-prem CDSW as well as the integration work for cloud CDSW based on AWS tech stack.

• Hardened the Virtual File System's gRPC file streaming operations and fixed various resource leaks, symlink venerability, and set up the gRPC server/client tests from scratch.

• Unified the GoLang package management system across ~20 components with 'go module'. Reduced the build time to 1/3, and increased network stability by using internal proxy registry.

• Integrated CDSW SSO with internal identity provider (SAML)

• Implemented the feature flag to disable / enable file upload / download.

Tech stack: GoLang, Docker, Kubernetes, AWS, gRPC, Helm, PostgresSQL, Node.js

Built an Analyzer backend service for VM instance recommendation as part of the proof-of-concept

development. (https://github.com/Hyperpilotio/analyzer)

• Deployed containers to cluster and benchmarked system/service level metrics under workloads.

• Implemented the recommendation searching strategy and visualized the behavior of ML algorithm.

• Maintained Analyzer service API that communicates with front-end and load test service.

This project involves working with Django, MongoDB, Docker, Kubernetes, Intel-snap, Kafka, SciPy.

Developed a musical onset detection algorithm and improved the performance by 7%. The work was

published to IEEE Signal Processing Letters, 2015. (https://github.com/cheyuanl/OnsetDetectorCLR)

• Researched audio signal reconstruction error using constrained (sparsity/non-negativity) least squares.