Designed and led an HCI study investigating how AI feedback timing affects emotional support in journaling. MyLova asks users to write about a negative event, then delivers GPT-4o-generated feedback — an empathic acknowledgment, a perspective-shifting reframe, and a practical suggestion — either almost immediately (~3 sec) or after a one-hour delay, to test whether the timing of that feedback changes how supportive it feels.

Built the Flutter/GPT-4o app from scratch and ran a four-week between-subjects field study (N=24, 549 diary entries) comparing the Immediate and Delayed conditions, using in-app logs and post-study interviews.

Overall willingness to accept the feedback and perceived AI understanding didn't differ significantly between conditions. But in the Delayed condition, participants' affective state at the moment feedback arrived was positively associated with both outcomes — suggesting delay's value depends on whether the emotional need is still present when the feedback lands.

Interviews suggested that delay can help users emotionally distance themselves, calm down, and reflect more deeply, but may also feel less relevant once the original emotional need has faded.

First-authored "Now or Later? Exploring Feedback Delivery Timing in AI-Based Emotional Support," accepted at ACM UbiComp'26.