At Google, I build production data-analytics agents on Gemini and ADK that turn open-ended questions into cited, chart-supported insights. I designed and shipped an agentic RAG pipeline for schema and business-context retrieval, reducing schema-linking errors compared with a single-shot RAG baseline.
I own the RLxF workstream for NL2SQL, where I built preference-pair pipelines, trained reward models, and ran preference-optimization fine-tuning that improved execution accuracy over the SFT baseline. I also extended post-training to multi-step agent behavior and defined the evaluation methodology used to gate launches.
At Twitter, I worked across personalization, retention, and recommended push notifications. The retention initiative I helped build drove 15M+ incremental monetizable DAU, and recommended push notifications grew to more than 10% of Twitter’s total monetizable DAU.
I also contributed to research published at venues including CVPR, ICML, and VLDB; my publications have received 2,900+ citations. I bring together hands-on model and agent development with experience shaping technical direction across large-scale ML systems.
