ChunkIt: an AI-driven planning tool that breaks overwhelming tasks into manageable, scheduled steps. Our first application: young adults with ADHD.


We ran semi-structured interviews with 10 participants with ADHD, aged 20 to 27. They included analysts, software engineers, a consultant, a marketer and master's students. We used thematic analysis to code their responses. Key insights:
These became five design goals: simplicity, personalization, a balance of structure and flexibility, AI-driven assistance, and visual, interactive features.
Users enter a task and its final due date. A few-shot-prompted LLM (Gemini 1.5 Flash) uses labelled chunking examples to split the task into subtasks, each with its own intermediate due date. After a Google sign-in, each chunk can be added to the user's Google Calendar through the Calendar API, which triggers push reminders.



In a within-subjects study, 7 participants got a 2-minute overview, then thought aloud while chunkifying 3 real tasks and adding every subtask to their calendar. A 15-minute interview followed, and the same interview was repeated after one week. Tasks included study-abroad prep, moving out of college, grad school applications, a pitch for a modeling system and an album-release plan.
Most participants finished every subtask they scheduled through ChunkIt during the week.
"Even if you procrastinate until the due date, you're only doing it for one task. By subtask 2 or 3, you feel accomplished." (P3)
"If it's not in my calendar, it's not getting done." (P3)
Most participants preferred 1 to 3 reminders per task. The benefit depended on already using Google Calendar, and some participants asked for Notion integration.
Participants asked for finer-grained chunks for complex tasks, control over the prompt, a "nudge" feature, dynamic rescheduling after missed deadlines, energy-aware scheduling and positive reinforcement.