Software Engineer Graduate, Data Ecosystem )

About the Role

The TikTok Data Ecosystem Team is responsible for crafting and implementing a storage solution for offline data in TikTok's recommendation system, serving over a billion users. The team's primary objectives include ensuring system reliability, uninterrupted service, and seamless performance. They aim to build a storage and computing infrastructure that can adapt to various data sources within the recommendation system, accommodating diverse storage needs. The ultimate goal is to deliver efficient, affordable data storage with easy-to-use data management tools for recommendation, search, and advertising functions. Talented individuals are sought to join the team in 2026. Successful candidates must commit to an onboarding date by the end of 2026.

Key Responsibilities

• Design and implement real-time and offline data architecture for large-scale recommendation systems.
• Build scalable and high-performance streaming Lakehouse systems that power feature pipelines, model training, and real-time inference.
• Collaborate with ML platform teams to support PyTorch-based model training workflows and design efficient data formats and access patterns for large-scale samples and features.
• Own core components of our distributed storage and processing stack, from file format to stream compaction to metadata management.

Requirements & Tech Stack

• Individuals completing or recently completed a PhD degree in Software Development, Computer Science, Computer Engineering, or a related technical discipline.
• Experience building large-scale distributed systems, preferably in storage, stream processing, or ML infrastructure.
• Solid understanding of Apache Flink internals, with hands-on experience in state management, connectors, or UDFs.
• Familiarity with modern Lakehouse technologies such as Apache Paimon, Iceberg, Delta Lake, or Hudi, especially around incremental ingestion, schema evolution, and snapshot isolation.
• Proficiency in Java/Scala/C++, and strong debugging/performance tuning ability.

Preferred Qualifications

• Experience in designing and optimizing Flink + Paimon architectures for unified batch/stream processing.
• Familiarity with feature storage and training data pipelines, and their integration with PyTorch, especially for large-scale model training.
• Knowledge of columnar file formats (Parquet, ORC, Lance) and how they are used in feature engineering or ML data loading.
• Previous experience in Lakehouse metadata management, compaction scheduling, or data versioning is a plus.
• (Optional) Knowledge of legacy data stores like HBase/Kudu is a bonus but not required.