Overview
A Slack-based predictive assistant that detects early warning signals of laptop failure and surfaces plain-language root causes with guided resolution paths.
Role
Project Lead
Design PM
UX Research
Front-end dev
Team
Kira B.
Laura K.
Lexeigh K.
Thea B.
Timeline
Jan - Jun 2025
11 weeks
Sponsors
Amazon IT
Mike & Rolf
| CONTEXT
A Blue Screen of Death rarely happens out of nowhere.
Most system crashes are preceded by measurable signals. Rising boot times, elevated memory pressure, degraded battery cycles… but users only see symptoms: "my laptop froze," "it just went black."
How might we detect early failure signals and intervene before disruption occurs?
| GOAL
Reactive to Proactive IT
Traditional IT support is reactive: crash happens, ticket opens, replacement ships. By then, unsaved work is gone and productivity is lost.
The goal of our capstone is to shift enterprise IT support from reactive repair to predictive prevention. Translate raw device telemetry into plain-language guidance that non-technical users can act on — and route them to the right resolution path (self-fix, backup & replace, or reassurance) without wasting time on misdiagnosed tickets.
| PROCESS
Process As of April.9
As a team, we trained a random forest model on 90th-percentile telemetry data across 12 device features (memory utilization, boot time, CPU capacity, battery cycles) to predict imminent system failures. Scoped the backend as on-demand diagnostics to avoid infrastructure complexity, and built a Slack chatbot that delivers proactive alerts via DM with ranked root causes and confidence scores.
My contributions:
End-to-End UX Research to Strategy
Defined the full interaction framework from alert → triage → resolution across all severity tiers, worked backwards from the customer research I lead in the past two months.
Platform Pivot
Led the shift from kiosk & webapp to Slack-native experience after identifying accessibility gap in end user's troubleshooting scenarios.
Problem Reframing
Shifted scope from general IT troubleshooting to predicting imminent crashes — sharpening every design decision downstream.
Data Output to Front End Flows
Turned confidence scores and feature importances into language and flows non-technical users could act on and provide feedback during our upcoming usability testing sessions.
| KEY ITERATIONS
01. Form Factor Challenge
Desktop app is inaccessible if your laptop is crashing / unresponsive.
Kiosk is inaccessible to remote workers; it can be awkward to troubleshoot laptop in the middle of a hallway :0
Previous Kiosk Form factor
| TAKEAWAY
Data + interpretability = value
A confidence score means nothing if a user can't map it to an action.
UX must translate system signals into human decisions.
The hardest design problem was how to be proactive in a helpful non- creepy way.
Delegate Ownership
As much as I love this project, owning everything will lead to burnout. I learned to trust my team and it had been amazing!
| NEXT STEPS
Next Steps
01
Validate predictive thresholds against real-world telemetry signals
02
Design a feedback loop between user actions and model confidence updates.
03
Explore event-triggered alert integration with existing monitoring infrastructure
04
Evaluate impact on IT ticket volume and average resolution time post-deployment