Guibin Zhang

Built and shipped production systems end to end across autonomous agents, ML ranking/model serving, and distributed compute. Designed and deployed a LangGraph-based autonomous, self-healing agent platform for near-real-time diagnosis and automated remediation of data-governance failures, improving dataset coverage and operational reliability. Owned next-generation agentic offline privacy compliance architecture, automating governance across large offline datasets with high reliability and faster coverage velocity. Implemented LightGBM-based ads preliminary filtering (replacing a rule-based negative cache) and delivered an offline-to-online evaluation pipeline to accelerate feature iteration and A/B rollouts. Re-architected Meta’s Spark driver for ML training/pipelines, reducing job failures and driver instability while enabling high concurrency and large-scale execution.

Built and scaled Workday Prism Analytics components, including a customer-facing Prism Lens Builder engine that turns petabyte-scale datasets into pre-aggregated lenses for sub-second queries. Designed and shipped an asynchronous Priority Batch Scheduler for batch jobs and its scheduling primitives used as a foundation for a Sync Query Scheduler. Optimized Spark Catalyst window functions across Spark 2.4 and 3.1 to eliminate redundant shuffles and improve runtime for row-level analytical workloads.

Built a production ML platform from scratch on Apache Spark and Scala for real-time bidding optimization. Designed the core bidding ML engine and a fault-tolerant, high-throughput model-serving pipeline, decommissioning legacy infrastructure. Developed CI/CD and release automation across QE, Build, and Ops environments, enabling zero-downtime deployments and reducing release-cycle time. Led technical execution and mentored engineers on Spark internals, Scala patterns, and ML platform best practices.