Sep 23 Meeting Log - snuhcs-course/swpp-2026-project-team-18 GitHub Wiki
Agenda: Review TA Feedback for the JustInTime Proposal
-
Differentiation from similar apps (e.g., Arrive)
- JustInTime includes a risk-tolerance parameter, enabling personalized learning and recommendations.
- It supports public transportation–related features.
- It appears to reflect real-time changes, unlike comparable apps that may not provide this capability.
-
Clarify the arrival criterion
- A user is considered to have arrived at the destination when they enter the destination geofence and remain there for at least two minutes.
-
Prepare the learning method and cold-start strategy
- Preparation time: Start with a default of 30 minutes. This default may be adjusted based on demographic factors such as age group and gender.
- During initial setup, collect user input about detailed preparation plans.
- Learn from the time between when the user dismisses the alarm and when they leave home, then adjust the estimated preparation time accordingly.
- The proportions of individual preparation tasks can be adjusted rather than requiring users to repeatedly enter detailed plans.
- Provide a lightweight feedback UI for finer adjustment, while designing incentives so users do not find feedback burdensome.
- Learning approach: Consider Bayesian updating or fitting a normal distribution. Based on a selected risk parameter, (\tau), the system can estimate an appropriate value from the distribution and provide a personalized recommendation.
- Preparation time: Start with a default of 30 minutes. This default may be adjusted based on demographic factors such as age group and gender.
-
Strengthen the testing plan
- Compare the learning-based model against a baseline model with a fixed buffer.
- Evaluate performance using:
- Number of late arrivals
- Difference between actual and target arrival times
- Test reliability under challenging conditions, including:
- Device lock state
- Battery-saving mode
- Device reboot
- Network errors