LLM-Fueled Data Acquisition

Led the full rebuild of Houzz’s Search Ecosystem V2, delivering major improvements across: 🔍 Search Ranking & Query Understanding: Designed and implemented core modules like query rewrite, redirection, autocomplete, trending search, and vertical intent detection, achieving consistent double-digit metric gains. 📊 End-to-End ML Platform: Identified the need for better feature and data pipelines, leading the redesign and redevelopment of the entire lifecycle—from raw data logging and feature engineering to model training (XGBoost, BERT, PyTorch, TensorFlow) and deployment of scalable, high-performance backend services primarily in Java/Python. 🎯 Personalization & Cold Start: Built a personalized ranking system tailored to audience segments, and deployed a reinforcement learning (multi-armed bandit) framework to solve the cold start problem for new/private label products. 🤝 Cross-Functional Leadership: Collaborated closely with product, data, and infra teams to align ranking objectives with business goals, driving substantial revenue and margin improvements. 🧠 Deep Learning Integration: Guided a team effort to adopt open-source DL models for vertical intent ranking, leading to significant revenue lifts. Skills: Hadoop · Python · Problem Solving · PHP · High Performance Computing (HPC) · PyTorch · SQL · Go (Programming Language) · XGBoost · System Performance · TensorFlow · Java

🌐 Federated Search Ranking: Designed and launched a global federated search ranking platform with vertical intent modeling across 13 international markets. Initial rollouts in Taiwan and Hong Kong showed strong engagement gains, expanding the team from 1 to 5 engineers. 🚀 Low-Latency Serving Platform: Built a high-throughput, low-latency system to serve machine-learned ranking models at scale. 🧱 Data Infrastructure: Developed a robust ETL pipeline and feature store capable of processing trillions of records, powering ML training and experimentation. 📈 Impact: Platform improvements led to a 25.3% increase in CTR and significantly boosted user engagement. 📊 Advertisement Data Warehouse: Led development of closed-loop attribution and ad exposure reporting systems.

🏥 Nurse Information System: Designed and implemented a user-friendly web-based system to help nurses manage patient care data. Built the entire architecture using Spring, Struts2, iBatis, jQuery, and MySQL. 📄 Electronic Medical Records (EMR): Developed Java and .NET libraries to convert legacy EMRs into CDA R2 (HL7v3 standard). Built internal HDFS-based storage with HTTP API support for scalable EMR ingestion.

🏭 Manufacturing Execution System (MES): Built middleware on BEA WebLogic to control wafer probing machines, track customer products, and connect MES with ERP systems. Reduced client-database communication issues significantly. 🌀 Central Wafer System: Brought modern IT practices like connection pooling and microservices to normalize 300+ proprietary wafer formats into a standard schema, resolving data quality issues and improving system scalability.