Chenming Ge

Chenming Ge

he/him

HCI · Generative UI · Computer-use Agents

Revision-Based Learning from Interactive Experience and Feedback

Submitted to NAACL 2027, 2026

RELIEF's four-step workflow: generate an interface, collect feedback, revise the interface, and fine-tune the generation model using revised traces.
RELIEF learns from feedback-guided interface revisions. Method overview from Figure 1 of the paper.

Jiawen Li, Chenming Ge, Alice Gao, Rishab Vempati, Yi-Hao Peng, Jason Wu
Manuscript, 2026. Submitted to NAACL 2027.

RELIEF turns feedback-guided revisions of generated interfaces into training supervision for code generation models. We collected 355 interaction recordings and accompanying feedback from 24 users across 200 generated websites; critiques inferred from human interaction recordings led to the largest observed usability improvements across the two models tested.

Paper (PDF) · Project Page · Research Experience