Hi, I'm Shubham.
I'm a Senior Applied Scientist, Tech Lead at Amazon, where I set technical direction for a 20-scientist research organization building foundation LLMs, agentic AI systems, and reinforcement learning pipelines for Amazon Nova and AWS AI Agents.
With 12+ years of experience, I bridge cutting-edge research and production-grade AI systems — personally running RL post-training experiments (PPO, DPO, DMPO, GRPO) on the production Nova model while mentoring scientists across the full LLM lifecycle, from architecture and post-training to reasoning, planning, and autonomous agents.
I write about LLMs, agents, and lessons learned on my blog. Check out my publications or get in touch.
Highlights
Amazon Nova
Setting technical direction for coding-agent post-training, contributing to Top-5 SWE-Bench performance
Technical Strategy
Sets technical strategy and mentors a 20-scientist organization across data, training, RL, and evaluation
Generative AI Agents
Building next-generation autonomous AI systems and agentic pipelines
Evaluation Infrastructure
Built and scaled an internal evaluation platform onboarding public benchmarks (SWE-Bench, BFCL, Tau-Bench) and a gym-authoring platform used by 30+ scientists and engineers
10+ Papers
Published at NeurIPS, ACL, EMNLP, CoRL, ECML
Leadership Approach
Strategic Thinking
Defining multi-year AI research and product roadmaps aligned to business outcomes
Cross-functional Leadership
Guiding scientists, engineers, and product teams across data, training, RL, and evaluation
Research → Production
Translating cutting-edge research into scalable, production-grade systems serving millions
Technical Depth
Hands-on involvement in model architecture, post-training, and agent system design