Work
Engineering Leadership at the Intersection of AI and Innovation
Over 15+ years across industry and academia, I’ve worked on AI platforms, agent systems, evaluation, NLP, and applied research — from hands-on research to building teams and setting technical strategy.
- Responsible for 12 people across AI Platform, AI Quality, and Data Science, leading the teams behind AI infrastructure, agent quality, evaluation, and data capabilities.
- Own the agent platform: the LLM gateway, Temporal-based agent orchestration, LangSmith observability, document processing and search, composable agent frameworks, and core AI products including AI Chat, Annotator, Request Analysis, and AI Actions.
- Established an eval-first culture across the AI organization, building offline and online evaluation, regression testing, and production LLM-as-a-judge monitoring from the ground up.
- Improved the scalability and reliability of the agent platform, including LLM orchestration, document processing, with a focus on observability, alerting, and incident response.
- Own AI cost visibility and optimization, with cost attribution and reduction plans across agents, LLM usage, document processing, and infrastructure.
- Primary technical contact for Anthropic, OpenAI, and Google Vertex AI; helped hire around 20 engineers and built the Data Science hiring process end to end.
- Helped define the AI strategy and roadmap, including agent orchestration, evaluation, and data strategy.
- Built hiring and onboarding processes for the AI org, including interview loops, evaluation criteria, and onboarding plans.
- Founded and led six-person team to build and launch Lumiro, an AI-powered tour guide that delivers interactive audio experiences for travelers.
- Owned end-to-end product and engineering execution across iOS, Android, backend services, and cloud infrastructure.
- Shipped core AI experiences including personalized tour generation, photo-based landmark discovery, and shared group tours.
- Led go-to-market strategy across pricing, customer acquisition, partnerships, and retention.
- Led the development of a new team to build and maintain the speech engine for Twilio Autopilot.
- Reduced the incidents and on-call duties of the GA product by 80%.
- Led the End Of Life effort of Twilio Autopilot: owned the whole process from planning to execution, collaborated effectively with the leaders in the company from Product to Finance, PS to GTM, Sales Enablement to Developer Experience, communicated with the customers for a smooth migration.
- Developed a rule-free, fully generalizable, T5-based classify-extract-normalize model to recognize the receipt numbers and improved the F1 score of the previous model by 35%, led a first big customer to sign a contract
- Designed and built a Machine Learning Metadata service to address the shared need for principled registration, promotion, and rollback of machine learning models and servables across different environments in a reliable and traceable manner.
- Designed and led Asset Manager Service that addresses the lifecycle of audio and text assets at Twilio AI Team.
- Designed and built Transformer Service that runs on Kubernetes, which can be used to train highly accurate and light models with a few examples.
- Worked on finding better representations using neural networks for processes such as tobacco smoking and UV light that result in cancer.
- Developed a benchmarking tool for somatic variant caller analysis