• Worked in the team building and maintaining Databricks Notebooks (Collaborative coding environment with Jupyter protocol, 500M ARR with 30% YOY growth) with TypeScript, React on the frontend, Scala, Python, Spark on the backend, and iPyKernel (Python), Almond (Scala) on the REPL side. Special focus on Notebook's Spark integration, and customers' code running environment (i.e. REPL and kernel-tending process)

• Architected, built, and maintained Spark Streaming DataFrame UX in Databricks Notebooks for Serverless and Standard (formerly Shared) mode clusters, unblocking critical customers like JP Morgan Chase and OpenAI, and fully integrated the 321M ARR product

• Spearheaded in landing Databricks' Serverless GA mechanism, implemented and verified critical CUJs for onboarding under tight time constraints. This effort readied the announcement for Serverless in DAIS 2024, which now has 46 Million ARR that is still rapidly (60% QOQ) growing

• Designed, built, and maintained multi-language features on Databricks Notebooks, notably Run Selected Text (8,000 WAU and 60,000+ requests/day), Python Formatter (7,400+ WAU), secure low-privileged Python library installation (50K+ DAU). Overwhelmingly positive customer feedbacks were received

• Hardened team technical structure with a multi-pronged approach: Runbot job for feature parity validation, detecting and triaging 2343 test failures, deflaked integration tests, and removed 42 suites from regular sign-off tickets, designing and developing generalized M3-based metric safeguarding Serverless feature development.

• Designed, created, and deployed RAG LLM Chatbot to reduce participating teams’ on-call load by answering internal technical questions, facilitated by Slack knowledge vector database and Wiki pages

• Designed, implemented, and maintained LLM-enhanced data ingestion pipeline for internal Slack messages, facilitating message analysis dashboards for oncall-heavy engineering teams across Databricks

• Integrated Amazon Chime services to Prelude, an internal platform facilitating meeting scheduling between recruiters and applicants written in TypeScript, improving user experience for 2200+ AWS recruiters and 10x more applicants each year.

• Upgraded Prelude user and customer's website layout with React and Redux framework, enabling Chime Meeting generation and association together with automatic email notification (with AWS Lambda and Kinesis)

• Designed (with Smithy), implemented and tested (with Jest) the APIs and the corresponding HTTP calls to make the abovementioned workflow possible

• Boosted Zhihu search engine's query rewrite capability with Google's Transformer deep-learning model implemented in TensorFlow 2.0, pushing the recall quality for long-tail keywords to a higher level

• Developed a query evaluation tool with GPT-2 language model for misspelling correction, which works together with above to better support more than 10,000,000 queries on Zhihu each day