Looking for a job

Mark Liu

@markliu2

Senior Machine Learning Engineer at DoorDash building GraphSAGE restaurant discovery and semantic search, with 80% lower LLM feature-generation costs.

United States

Message

At DoorDash, I build graph-based restaurant discovery and semantic search for recommendation workloads. I designed GraphSAGE models and candidate-generation jobs to improve discovery for restaurants with limited interaction history.

I also helped design LLM-assisted menu catalog enrichment and connected semantic attributes with consumer interactions for ranking. By evaluating quality and cost trade-offs, caching prompts, and materializing features on events, I reduced LLM feature-generation costs by 80%.

At Netflix, I worked on content personalization, from sequence-based recommendations to candidate ranking and serving. I built offline ranking evaluations that measured a 7% relative improvement in NDCG@10, and reduced p95 inference latency by 23% through serving optimizations.

Experience

Work history, roles, and key accomplishments

Current

Senior Recommendation Systems Engineer

Current

DoorDash

Apr 2023 - Present (3 years 7 months)

Designed graph-based restaurant discovery and semantic search capabilities, and built streaming features and inference services for recommendation workloads. Reduced LLM feature-generation computation costs by 80% and supported a 1.8% relative lift in push and notification-hub engagement.

GraphSAGE PyTorch Geometric GNN Apache Flink Kafka Snowflake Elasticsearch Redis LLM VLLM Kubernetes Docker A B Testing

Senior Machine Learning Engineer

Netflix

Sep 2019 - Mar 2023 (3 years 6 months)

Worked on content personalization, including sequence-based recommendation, candidate ranking, offline experiments, and low-latency serving. Achieved a 7% relative improvement in NDCG@10 and reduced p95 inference latency by 23%.

Multi VAE SASRec TiSASRec PyTorch Java Vector API Doubly Robust Estimators Contextual Bandits Distributed Training Model Monitoring

Senior Recommendation Systems Engineer

Accenture

Jan 2017 - Aug 2019 (2 years 7 months)

Delivered retail recommendation and forecasting solutions, including Spark feature pipelines, model deployment automation, and production monitoring. Improved recommendation Precision@10 by 8% and cut model deployment cycle by 30%.

MLlib Databricks MLFlow Docker Kubernetes Collaborative Filtering Demand Forecasting Data Quality

Recommendation Systems Engineer

Oracle

Jun 2014 - Dec 2016 (2 years 6 months)

Built Hadoop/Spark data workflows and batch collaborative-filtering jobs for enterprise personalization. Reduced batch processing time by 27% and improved Recall@10 by 7%.

Hadoop Spark Hive HDFS MLlib Oracle Big Data Discovery Collaborative Filtering Data Quality