My path into AI started with a deep fascination for how machines learn from data. Early in my career, I built ML models for NLP and information retrieval problems, learning to navigate the gap between research benchmarks and real-world complexity. That tension—between academic rigor and engineering pragmatism—has defined my approach ever since.
Over 12+ years, I evolved from applied researcher to technical lead, shipping production systems used by millions across Alexa, AWS, and enterprise GenAI platforms. Along the way I took on more technical direction: setting strategy, building research culture, and translating long-horizon bets into near-term milestones, while staying personally hands-on with the hardest problems.
Today at Amazon, I set technical direction for a 20-scientist research organization focused on coding-agent post-training within Amazon Nova, AWS's foundation model family, and AWS AI Agents. My work spans the full LLM lifecycle: architecture decisions, post-training (SFT, RLHF, GRPO), reasoning and planning, and autonomous agentic systems — and I still personally run RL post-training experiments and built the team's internal evaluation infrastructure. I collaborate with scientists, engineers, and product leaders to ship AI that is both frontier-capable and production-reliable.
I'm a published researcher at ACL, EMNLP, NeurIPS, CoRL, and ECML, and I believe that staying close to the research frontier is what keeps engineering decisions grounded in what is actually possible—not just what is currently deployed.
Amazon.com Services Inc. · Sep 2017 - Present
Sets technical direction and mentors a 20-scientist research organization focused on coding-agent post-training for the Amazon Nova foundation models and AWS AI Agents. Personally runs PPO/DPO/DMPO/GRPO post-training experiments on the production Nova model. Built and scaled an internal evaluation platform onboarding public benchmarks (SWE-Bench, BFCL, Tau-Bench, ScienceAgentBench, Terminal-Bench), and a reusable RL gym-authoring platform used by 30+ scientists and engineers to scale to 100+ RL environments.
Amazon (Alexa AI) · Earlier roles, same Amazon tenure
Led Alexa Conversations internationalization across 16 locales, driving 6B+ annual conversations and 200% adoption growth. Delivered key launch features improving interactions for 100M+ customers.
Great AI leadership is about setting a direction that outlasts any individual contribution. I invest heavily in defining multi-year research roadmaps that align to real business outcomes, while leaving space for the discoveries that only emerge through experimentation.
The best research cultures are built on psychological safety, clear ownership, and a shared sense of mission. I try to create environments where scientists feel free to take risks, challenge assumptions, and grow faster than they would anywhere else.
Translating a research idea into a production system that serves millions of users requires a different kind of rigor than publishing a paper. I have learned to build that bridge deliberately—through careful system design, robust evaluation, and tight feedback loops between research and engineering.
I believe leaders in AI must remain technically grounded. I stay involved in architecture reviews, read papers regularly, and maintain hands-on familiarity with the systems my teams build. Long-term thinking in AI requires knowing what is on the frontier today.