Senior ML Infra Architect — Real-Time, Large-Scale ML

Roblox; San Mateo (CA); On-site; USD 243,000 - 295,000; Full time

Job summary

Roblox is seeking a Senior Machine Learning Infrastructure Engineer in San Mateo, CA to build scalable ML infrastructure that powers matchmaking, recommendations, and search. You will work across model architecture and data systems with ML engineers and data scientists to push the boundaries of large-scale training and serving.

You will design streaming pipelines, online learning, and low-latency inference, ensuring data correctness and reliable deployment.

Qualifications

• 5+ years designing, building, and deploying large-scale ML/data systems in production.
• 3+ years leading ML infra or data engineering teams.
• Hands-on with streaming/data processing (Kafka, Flink, Spark, Beam).
• Experience with online/continual learning for recommender, ranking, or search is a strong plus.
• Strong communication and collaboration across teams.

Responsibilities

• Design and own large-scale data engineering and streaming systems for ML.
• Build real-time event ingestion, streaming feature computation, and data quality guarantees.
• Develop online and continual learning systems with model updates and safe deployment.
• Evaluate tradeoffs across streaming, training, and inference for low-latency serving.
• Collaborate with data scientists, ML engineers, and product teams to deploy robust ML platforms.

Skills

ML infrastructure

Data systems design

Streaming systems

Low-latency storage

Communication skills

Leadership

Education

BS/MS/PhD in CS/Engineering

Tools

Kafka

Flink

Spark

Beam

Redis

Feast

Tecton

Cassandra

Job description

Roblox is seeking a Senior Machine Learning Infrastructure Engineer in San Mateo, CA to build scalable ML infrastructure that powers matchmaking, recommendations, and search. You will work across model architecture and data systems with ML engineers and data scientists to push the boundaries of large-scale training and serving.

You will design streaming pipelines, online learning, and low-latency inference, ensuring data correctness and reliable deployment.