Jesse Steinweg-Woods
Staff Machine Learning Engineer @Change.org
Signup · Get unlimited contacts
WORK HISTORY
Staff Machine Learning Engineer @Change.org
Monterey Park, CA, US
Implemented new petition recommender system, increasing sign rate by 30-50%(US had largest lift among international markets), served to 50 million users/week- Recommender system is based on two-tower retrieval with pointwise ranking. Both models utilize a transformer architecture on signature sequences (inspired by Chen et al. 2019 from Alibaba) with DCN V2 for feature crosses, trained with PyTorch- Used Ray on EKS for distributed training and inference, with Airflow orchestrating the jobs- Incorporated A/B testing framework with feature flags to allow easier testing of new features/model architectures while maintaining weekly fine tuning to keep the models fresh- Retrieval model incorporates batch negative sampling with log-Q correction for faster training/better performance- See attached article AWS wrote about the system for more details
EDUCATION
Penn State University
Bachelor of Science (B.S.), Meteorology
Texas A&M University
Doctor of Philosophy (Ph.D.), Atmospheric Science
Colorado State University
Master of Science (M.S.), Atmospheric Science
Penn State University
Bachelor of Science (B.S.), Energy, Business, and Finance
SKILLS
ABOUT JESSE STEINWEG-WOODS
I am a Machine Learning Engineer focused primarily on scalable recommender systems and search. My experience includes- recommender systems (two-tower, transformers, DCN V2)- search (embedding-based retrieval/RAG/learning to rank/Elasticsearch)- working with LLMs (encoder representations for search/recsys application)- distributed training/inference (Ray)- text classification- deployment of ML models in batch/streaming/online applications- integrating automated model retraining pipelines via Airflow- CI/CD for automated testing and Docker/Kubernetes deployments- AWS/GCP for cloud tools- Databricks (Spark) for data preparation/analysis
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.