Yuelin Zhang

Model Inference

ML Serving (2024/01 - current) Building scalable, fault-tolerant ML serving systems at Pinterest. Focusing on modern compute platforms (K8s) for ModelServing, next-gen feature serving, GPU optimizations, and multi-tenant large-scale inference infrastructure Ads Indexing (2022/05-2023/12) Focused on realtime processing (Flink) and batching processing (Spark, MapReduce) for Ads Indexes Trust & Safety (2019/04-2022/05) Built internal platforms to keep Pinterest context safe - focused on human review tools, automated triage systems, and anti-spam work. Keep Pinners Safe!

Team: AWS Deep Learning Project: Machine Learning Based GitHub Bot 1. Established an Email Bot which sends daily GitHub issue reports using AWS Simple Email Service and AWS Lambda. Wrote a python crawler using Breadth First Search algorithm. Retrieved more than 6k issues per execution. 2. Built a label recommendation service served on an AWS Elastic Beanstalk application. Implemented backend with Flask in a Docker container. Applied Natural Language Processing (TFIDF, Word2Vec, fastText) and Machine Learning (SVC, Multi-layer perceptron, Naïve Bayer) methods to do multi-label text classification. 75% predictions meet requirements. 3. Designed a Label Bot to realize automatic labeling. Built the bot on an AWS Lambda function with an Amazon CloudWatch event trigger. Stored credentials in AWS Secrets Manager. Realized automatic deployment using Terraform and Serverless framework. Decreased issue response time from 10 hours to 5 minutes.