Boran Han

Senior Research Scientist / Amazon AGI Foundations / AWS

Boran Han.

Building foundation models and learning systems for agentic tasks, structured data, and multimodal reasoning.

View research Google Scholar
Santa Clara, California / 37.35 N

Research focus / 01

Foundation models and reasoning systems for agents, structured data, and science.

ReSkill framework for creating and testing reusable skills during agentic reinforcement learning

Agentic RL / 2026

Skills that evolve with the policy

Reconciling reusable skill creation with reward-driven policy optimization.

SenTSR-Bench knowledge injection framework for time-series reasoning

Reasoning / 2026

Knowledge injection for time-series reasoning

Bridging specialist time-series models and general reasoning models.

Chronos-2 architecture for universal time-series forecasting

Foundation models / 2025

Universal zero-shot forecasting

One model for univariate, multivariate, and covariate-informed time series.

HetRL scheduling and execution framework for reinforcement learning across heterogeneous GPU environments

Systems / MLSys 2026

RL across heterogeneous hardware

Efficient LLM post-training across mixed GPUs, regions, and interconnects.

By the numbers / 02

Research metrics.

3K+ citations across machine learning, imaging, and natural science
29 selected publications in leading conferences and journals
149M+ total downloads for Chronos-2
11.7M+ combined total downloads across Mitra models

Recent work / 03

Recent publications.

When LLMs Read Tables Carelessly: Measuring and Reducing Data Referencing Errors

ACL Oral

LLMZero: Discovering Adaptive Training Strategies for RL Post-Training via LLM Agents

Preprint

ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL

Preprint

SenTSR-Bench: Thinking with Injected Knowledge for Time-Series Reasoning

AISTATS

HetRL: Efficient Reinforcement Learning for LLMs in Heterogeneous Environments

MLSys
View selected publications