Agent Improvement Lab
An independent public research project exploring reliable evaluation, failure diagnosis and controlled improvements for tool-using agents.
Ian Chen
AI research / evidence-led engineering
Better evaluation. More reliable AI agents.
I explore how agent evaluation, rubric quality and execution evidence can support systematic improvement. This portfolio connects research projects with technical writing, research plans and evaluation methodology.
Featured research
Research interests
From measurement to understanding
Agent evaluation
What makes an evaluation trustworthy when agents interact with tools, policies and changing state?
Rubric quality and reward signals
How can grading criteria measure the intended behavior and support useful learning signals?
Failure diagnosis and generalization
Which changes address observed failures, and do they transfer to untouched tasks?
Notes from learning and building
Transformer: Understanding Attention
A learning note on sequence models, encoder–decoder architectures and the attention mechanism behind Transformers.
Article in ChineseLLM Agents
An exploration of LLM-driven game characters, covering NPC behavior, interactions and the design of an agent system.
Article in ChineseResearch engineering, with evidence at the center
My interests connect LLM agent evaluation, rubric quality, reward modeling and reinforcement learning. This portfolio grows through independent projects and technical writing.
About →