Team Introduction:The TikTok Flink Ecosystem Team plays a critical role in delivering real-time computing capabilities to power TikTok’s massive-scale recommendation, search, and advertising systems. This team is focused on building the infrastructure for stream processing at exabyte scale — enabling ultra-low-latency, high-reliability, and cost-efficient real-time data transformations. We are deeply involved in developing and optimizing Apache Flink and surrounding components like connectors, state backends, and runtime execution models to meet TikTok’s rapidly evolving data needs at EB-level throughput and scale. We also collaborate closely with ML infrastructure teams to bridge real-time stream processing and machine learning. This includes integrating Velox to accelerate model training, building multimodal data pipelines, and utilizing frameworks like Ray to orchestrate large-scale distributed ML workflows.Responsibilities:

Design and develop core Flink operators, connectors, or runtime modules to support TikTok’s exabyte-scale real-time processing needs. - Build and maintain low-latency, high-throughput streaming pipelines powering online learning, recommendation, and ranking systems.
Collaborate with ML engineers to design end-to-end real-time ML pipelines, enabling efficient feature generation, training data streaming, and online inference.
Leverage Velox for compute-optimized ML data transformation and training acceleration on multimodal datasets (e.g., video, audio, and text).
Use Ray to coordinate distributed machine learning workflows and integrate real-time feature pipelines with ML model training/inference.
Optimize Flink job performance, diagnose bottlenecks, and deliver scalable solutions across EB-scale streaming workloads.

Machine Learning Engineer Intern (Training Pre-processing) - 2025 Summer (PhD)