Growth AI Applications: LLM Post Training, Agent, RecSys

Played a crucial part in user acquisition to reach millions of users worldwide. Scaling data platform, ML, and GenAI features to drive global growth:

• Led Pinterest's first GenAI production experiment on user activation and content enrichment, resulting in a significant 20% increase in traffic and a 60% boost in user engagement

• Identified and capitalized on opportunities to enhance the unauth user shopping initiative by building a shopping landing experience that prioritized trending searches, strong shopping intent, and showcased top product offerings. Achieved a 10% in quality sessions improvement and 80% shopping traffic growth.

• Designing and optimizing large-scale machine learning systems, including content ranking systems to spotlight Pinterest's top-tier content. This led to an 15% surge in traffic, 120M boost in WAU and highest content index rate from search engine in Pinterest history

• Designed and developed the Pinterest internal A/B testing experiment framework, which increased 30% engineers' weekly productivity across multiple teams

• Technologies: LLMs (fine-tuning, RAG), Machine Learning (PyTorch, Ray), Python, Scala, Java, SQL, Spark, Kubernetes, Airflow, AWS

• Developed a SmartRate API to select optimal shipping rates based on time-in-transit criteria, reducing costs by 15% for enterprise clients and processing over millions of requests daily with 99.99% uptime

• Engineered a high-throughput, fault-tolerant data pipeline using Kafka and Spark Streaming, processing 50TB+ daily shipping data with sub-second latency including live shipment tracking (200K events/sec peak)

• Optimized database performance for high-volume transaction processing: Implemented sharding and read replicas with efficient indexing strategies and query optimizations, reducing p99 query latency by 60%

• Technologies: Python, MySQL, Vertica, SQLite, Kafka, Docker, Jenkins, Vagrant

• ETL models to sort terabytes of customer feedback data

• Business category mapping system via DNN include Google, Facebook, Yelp, Bing and etc.

• NLP and sentimental prediction model to monitor for online reputation management utilizing to anticipate business needs and customer concerns.

• Technologies: Python (Pandas, NumPy, Keras, Scikit-learn, TensorFlow, Cherrypy, Flask), SQL, D3.js, Node.js, Maven, AWS(Redshift, S3, EC2, Mturk), Mongodb, Sumo Logic, Jenkins

• Actively participated in Edge Cloud Composites (ECC) research project

• Worked with a research team on integration of resource between personal devices/home appliances and personal clouds that created at hypervisor level