Charith Peris

Building Responsible AI
that works
at scale.

I lead applied science teams working at the frontier of large language models.

At Amazon's AGI organization, I lead research and production implementation in Responsible AI for foundation models and agents — from novel research to systems operating at Amazon scale. The threat surface in deployed AI keeps changing. My work is building safety infrastructure that holds even when the specific threat hasn't been named yet.

Earlier I built ML and data science systems in ad tech and healthcare start-ups, giving me a grounding in the high-stakes side of production AI. I hold a PhD in Physics from Northeastern University and was a predoctoral fellow at the Harvard-Smithsonian Center for Astrophysics.

Outside of work, I'm a choral conductor and singer: World Choir Games champion with Sri Lanka's The Revelations, and a former Harvard Glee Club member who performed at Carnegie Hall.

Research &
Impact

Publications and patents spanning LLMs, multilingual NLU, responsible AI, and probabilistic ML.

Selected Patents

US Patent 20210035163 · Feb 2021
US Patent 20180032914 · Feb 2018
US Patent 20180032913 · Feb 2018

Blogs

Writing on AI research, production ML, and the practical realities of building intelligent systems.

Multiagent AI for generating chain-of-thought training data
Amazon Science Blog · July 2025
Detoxification of large language models via regularized fine-tuning
Amazon Science Blog · November 2024
Continual learning in the federated-learning context
Amazon Science Blog · December 2023
Domain data trumps teacher knowledge for distilling NLU models
Amazon Science Blog · January 2023
Predicted Ghost Ads: An accurate and cost-effective method for measuring incrementality
Medium · DataXu