Dec 2024 – Present · 1 yr 3 mos · United States

• Co-teaching CS329A (Self-improving AI agents): https://cs329a.stanford.edu in Winter and Fall 2025
• Program Chair for MLSys 2026
• Research in self-improvement techniques for LLMs with RL and multi step reasoning

May 2024 – Dec 2024 · 7 mos

• Multimodal LLMs; Reasoning; Agentic LLMs
• synthetic data and SFT+RL for post-training to build Android control agents (improve reasoning and planning capabilities to control Android apps/screen similar to Computer use)
• efficient model architectures for inference efficiency

Apr 2023 – Apr 2024 · 1 yr · Mountain View, California, United States

• Gemini Models: Lead Researcher in scaling and stability of pretraining the dense and sparse LLMs.
• Recognized by three spot bonuses for the exceptional contributions and leadership.

Apr 2022 – Apr 2024 · 2 yrs

• ML:
• Technical Lead for training, infrastructure and finetuning for PaLM model series, including first author of PaLM 540B model paper (https://arxiv.org/abs/2204.02311) that was featured in Google I/O 2022.
• Co-led research initiatives on improving reasoning and multimodal capabilities of PaLM, LLM instruction tuning, as well as optimizing inference costs with 10+ research publications.
• Co-led domain adaptation projects of PaLM, such as MedPaLM for medical domain and PaLM-E for robotics domain.
• Recognized by Research PA Tech Impact Award in 2022 and 2023.
• Recognized by 15+ spot bonuses and 20 peer bonuses for leadership.
• Systems:
• Lead researcher in the system design and implementation of Pathways system at Google to enable large-scale ML training across multiple TPU v4 pods (https://arxiv.org/abs/2203.12533) at Google. Pathways systems paper recognized by Outstanding paper award in MLSys 2022.
• Novel algorithm design to reduce inference cost of large models (Efficiently scaling Transformer inference). Recognized by Outstanding paper award in MLSys 2023.

Sep 2020 – Apr 2022 · 1 yr 7 mos

• Core member of Pathways system, contributing to design and implementation. Led design of efficient AI algorithms using sparsity and multi-task learning in research. Top-tier NeurIPS publications including Sparse is Enough in Scaling Transformers and Dselect-k: Differentiable selection in the mixture of experts with applications to multi-task learning

Feb 2018 – Sep 2020 · 2 yrs 7 mos

• Led design and implementation of on-device optimization algorithms, including model quantization, for vision use cases on mobile devices and microprocessors. Core member of TensorFlow Lite team.

Oct 2015 – Jan 2018 · 2 yrs 3 mos · Greater New York City Area

• Novel distributed ML algorithms for video analytics.
• Led interdisciplinary team across machine learning and systems and networking.
• Co-supervised one graduate student, two undergraduate student theses, and two interns.

Oct 2013 – Sep 2015 · 1 yr 11 mos · Redmond, Washington

• Novel distributed ML algorithms to speed up large-scale machine-learning queries on videos.
• Led interdisciplinary team on edge computing for videos across machine learning and systems/networking.

Jun 2012 – Sep 2012 · 3 mos · Redmond, WA

Jun 2008 – Sep 2008 · 3 mos · Redmond, CA

Jan 2008 – Jan 2012 · 4 yrs

Sep 2007 – Aug 2013 · 5 yrs 11 mos

• Published 13 research papers on signal processing in multi-antenna systems for wireline & wireless broadband-access networks.
• Novel algorithms to enable Gbps Internet access speeds.
• Led standardization of the proposed models in technical standards of UK & US Access Networks.

May 2006 – Jul 2006 · 2 mos · Rennes Area, France