Present:Gemini: Generative AI and LLM for Google Search. Build the best GenAI platform in Google, and enable AI Overviews (formerly known as Search Generative Experience). Quality lead in Search AI Platform:* LLM end-to-end quality, content quality and auto raters* LLM modeling and LoRA adaptation * LLM prompting and prompt tuning * LLM agents and multi step reasonings* LLM query generation* LLM service and API infraPast:Tech lead manager of semantic document (web page) understanding and serving.Core search and knowledge engine quality. Semantic annotation of everything. Understand every query, page and beyond, in terms of and link them to Knowledge Graph. Understand users' intents. Provide the most accurate and relevant information to users.Worked on Google's system of query understanding and document understanding. This is a cornerstone system used in search and many other products across Google. Tech lead on semantic document annotation. Led the engineering team on quality and infrastructure to build a complex system from ground up, and apply it to various Google products.Tech lead for semantic model and annotator quality. Designed ultra compact index encoding for semantic data representation in Google search index. Seen through the life cycle of product development from ideas to production. Written highly optimized production code on cutting edge models and algorithms. Served as TL of production model release and feature launching gate keeper. Worked closely with various product areas in using the system across the fleet.

Elance merged with oDesk in Apr 2014 to Elance-oDesk, and re-branded as Upwork in May 2015.I continued working on Elance matching for a short period and then led a team of engineers to build a federated search engine for Upwork Messenger. This includeshierarchical internal data pipeline for both real-time and batch data processing.concurrent queries for multiple data sources, both internal and externalsearch grammar using Context Free Grammar parsercontinuous deployment for both the API and the indexer, including fully automatic data bootstrapping and zero downtime deploymentmaintained team’s own servers without opscomprehensive system metric collection, monitoring and alertingUpwork Messenger achieved a WoW growth of 8%.

Led the matching back-end services team to build cool applications with cutting edge technologies. * Multi-way matching (recommendation) engine, one single platform for all recommendation projects, supporting multiple data sources, multiple indexing storage, as well as easy-to-plugin matching plans (configurations), in a real-time manner. Built on top of Lucene/MongoDB, but with a bunch of stunning in-house features. * Health-check/app-metric/user feedback collection/relevance learning for matching engine* Geo API, a location service with a fruitful of features, supports hierarchical global data, name alias, fuzzy search, auto suggest, and IP mapping* Skill API, a WordNet like knowledge net, with similarities/hypernym/hyponym, auto suggest* Skill recommendation and budget estimation system for job posting, kNN based smart engine with self-revolution capability* A/B testing experiment system* End-to-end ETL pipeline, including MapReduce based data aggregation system for recommendations, Oracle Data Sync, and etc* A horizontally scalable and high through-put notification/email system with automatic failover.* Deployed and maintained daemons/cronjobs/service-apis in glassfish/tomcat, running in both data center and AWS* Led cross functional team, local and remote, for the effort of entire product life cycle* etc etc...Technologies and tools used:Lucene, SOLR, Elastic Search, Hadoop, MapReduce, Storm, Kafka, Zookeeper, Glassfish, Tomcat, Oracle, MongoDB, Java, Php/Symfony, Jenkins, Gerrit, Git, Perforce, Jira, Confluence, AWS