数据架构师（异构算力方向） - TikTok研发
Core
Design and manage the overall data architecture for heterogeneous computing resources to optimize AI training and inference costs, efficiency, and capacity planning for TikTok's large-scale model operations.
Role type
Senior Data Architect (Heterogeneous Computing & AI Infrastructure)
Builds
Unified metric systems, multi-layer data models for resource clusters, cost attribution tools, and capacity prediction models for GPU infrastructure.
Domain
AI Infrastructure / Heterogeneous Computing / Cloud Resource Management
Deliverable
production ML models
Required skills
Data warehouse design, Lakehouse architecture, Real-time computing, GPU infrastructure mechanics, Token lifecycle analysis, Cost attribution modeling, Capacity forecasting, SQL, Python
Preferred skills
Cross-team data system leadership, Data governance, Statistical analysis, Resource optimization strategies
Technologies
GPU clusters, Token metrics, KV Cache, Batching, Model routing, Data governance frameworks
Responsibilities
Plan the overall architecture for heterogeneous computing data covering data collection, modeling, metrics, and decision support; Design multi-layer data models from resource pools to business scenarios for minute-level monitoring; Build cost attribution capabilities to allocate GPU time and token consumption to specific models and projects; Identify resource inefficiencies and drive optimizations in scheduling and platform operations; Establish capacity prediction models to inform GPU procurement and budgeting; Define data governance standards for metrics, quality, and lineage.
Seniority
Senior, hands-on IC