Brian Lin

Built and shipped production GenAI shopping assistant and RAG platforms that integrated LLMs, embeddings, and tool calling to reduce manual planning time and improve retrieval accuracy; instrumented CI/CD, canary rollouts, and SLO-driven observability. Led search intent, query understanding, and ads ranking pipelines improving relevance and revenue metrics through robust evaluation and A/B testing

Developed production package-tracking extraction and near-real-time entity pipelines using TensorFlow, Kafka, Storm, and HBase, and maintained data refresh/backfill workflows with Spark and Hive to improve freshness and reduce false positives. Implemented regression tests, golden fixtures, and safe rollouts with feature flags and canaries.

Trained classifiers for Smart Views using TensorFlow and Spark MLlib, built DOM extraction services in Java, and orchestrated offline clustering and rule generation to serve low-latency lookup rules from HBase with precision/recall release gates.

Built text processing and sentiment analysis pipelines for FOMC transcripts using Python, pandas, NLTK, and scikit-learn; ran topic modeling and econometric event studies to produce reproducible datasets and analyses for research.

Built early warning models for part and vehicle failures using scikit-learn and XGBoost, joined warranty and repair datasets with Spark/Hive, and delivered risk-ranking dashboards and repeatable scoring pipelines for quality teams.

Activities and societies: Research assistant work on text processing and sentiment indices; coursework and projects in machine learning, NLP, and econometrics.

Completed a dual-focused undergraduate program covering computer science and statistics with coursework and research in text processing, machine learning, and econometric analysis.