Alexander Erik Huang Senior Machine Learning Engineer
************@*****.*** 650-***-**** Fremont, CA linkedin.com/in/erik-h-429869129 Senior Machine Learning Engineer with 10 years of experience building and deploying ML systems for ad ranking, recommendation, AI-powered search, and predictive analytics. Experienced in deep learning, two-tower retrieval, CTR/CVR prediction, LLMs, and RAG. Strong background in Python, SQL, and PyTorch, with hands-on experience optimizing production inference, evaluating model performance, running A/B experiments, and mentoring engineers. Skills
Programming
Python, SQL, JavaScript
Generative AI and Search
LLMs, RAG, NLP, Semantic Search, Multimodal Retrieval, Two-Tower Models, Embeddings, ANN Search
Cloud and Data Tools
AWS SageMaker, AWS Lambda, Apache Spark, PySpark,
Apache Airflow, MongoDB, Looker
Machine Learning
PyTorch, TensorFlow, Scikit-learn, XGBoost, Deep Learning, Recommendation Systems, Learning to Rank, CTR/CVR
Prediction, Model Calibration, Feature Engineering ML Engineering
Model Training, Model Serving, Distributed Inference, Inference Optimization, CI/CD, Model Monitoring, A/B Testing
Professional Experience
Senior Machine Learning Engineer
02/2020 – 09/2026 San Francisco, CA
•Developed personalized ad ranking models for Instagram Feed and Reels, combining user behavior, ad engagement, and contextual features to improve ad relevance and conversion prediction.
•Built two-tower retrieval models in PyTorch using user and ad embeddings, improving candidate Recall@K by 4% while maintaining efficient retrieval at scale.
•Improved CTR/CVR prediction through multi-task deep learning models incorporating user interaction sequences and conversion history, reducing normalized entropy by 2% compared with the previous baseline.
•Applied probability calibration to correct prediction bias across advertiser segments, improving alignment between predicted and observed conversion rates.
•Optimized candidate retrieval with hard-negative mining, approximate nearest neighbor search, and embedding caching, reducing p95 retrieval latency by 15% without significant loss in recall.
•Designed offline evaluations and A/B experiments to validate ranking improvements, with an estimated 1.5% lift in conversions while monitoring ad quality and user engagement.
•Developed LLM-powered query understanding and semantic retrieval components for Instagram search, improving discovery of relevant posts and videos through natural-language queries.
•Built RAG workflows combining text and multimodal content embeddings with retrieved context, achieving an estimated 6% improvement in NDCG@10 during offline search evaluation.
•Optimized search inference through precomputed content embeddings and request batching to reduce redundant computation and improve response latency.
•Mentored junior ML engineers on model architecture, experiment design, and production debugging while collaborating with Ads, Search, and ML Infrastructure teams on model deployment and reliability. BuildingConnected
Machine Learning Engineer
07/2018 – 02/2020 San Francisco, CA
•Improved NLP models for extracting project dates and construction details from emails, increasing entity extraction F1 by an estimated 5 percentage points through better preprocessing and training examples.
•Developed an XGBoost-based subcontractor recommendation model using project requirements, contractor profiles, and historical interaction features, with an estimated 7% improvement in Precision@10 over the existing ranking baseline.
•Built feature engineering and model training pipelines using Airflow, MongoDB, and PySpark to prepare datasets and support consistent recommendation model training.
•Deployed NLP and recommendation models using AWS SageMaker and Lambda, introducing CI/CD checks to automate testing and improve release reliability.
•Designed A/B experiments to evaluate recommendation quality and user engagement, and mentored an intern developing location extraction models for construction emails. Gobble
Data Scientist
11/2016 – 07/2018 Palo Alto, CA
•Developed customer churn prediction models using subscription history, ordering behavior, and engagement signals to identify at-risk customers and support retention campaigns.
•Built personalized meal recommendation models using customer preferences and ordering history to improve weekly menu selection.
•Developed ingredient demand forecasting models using historical orders and seasonal trends, with an estimated 10% reduction in forecast error compared with the previous forecasting approach.
•Deployed predictive models using AWS Lambda and developed Looker dashboards to track sales performance, customer retention, and purchasing trends.
Metis
Data Science Fellow
07/2016 – 11/2016 San Francisco, CA
•Built an image classification web application using Python, TensorFlow, and JavaScript, integrating trained models with a web interface for predictions.
•Developed regression and classification models for movie revenue forecasting and diabetes risk prediction, along with interactive clustering visualizations using D3.js. Education
Master's Degree in Statistics
University of California, Davis
08/2014 – 05/2016 Davis, CA
Bachelor's Degree in Mathematics
University of California, San Diego
08/2010 – 05/2014 San Diego, CA