How do you sift through billions of lines of database logs to isolate a single error code in less than 2 minutes?
Client
Amazon Prime Video
My Role
Sr. UX Designer / Solo
Timeline
6 Months
Design system
Enterprise UX
Complex solutions
Project overview:
The Problem
Following Amazon's acquisition of Thursday Night Football streaming rights, the company faced immense pressure to deliver a flawless multi-device viewer experience. The core design challenge was driven by a bold executive mandate: design an intuitive internal monitoring solution that allowed operations teams to detect, isolate, and resolve live streaming anomalies in under two minutes.
The Solution
Working cross-functionally, I led the UX initiative for a powerful diagnostic tool that allowed Prime Video technicians to pinpoint quality issues across billions of devices. By creating a dedicated design system and designing highly logical drill-down interactions, I empowered operators to visually filter global streaming data down to an exact, actionable problem code in real time.
"One software to rule them all"
Project Sauron was multiple applications that worked together to help its users make sense of large amounts of video analytics data.
Aggregate Filtering
Session Filtering
Anomaly Detection
Custom Dashboards
1
Design a robust filtering mechanism:
Designing an aggregate filtering system for massive datasets required an advanced, yet intuitive, structure. To achieve this, I created a solution that enabled users to build, save, and share complex filter presets. This streamlined their workflow, saving valuable time and allowing them to quickly isolate video playback error codes.
The aggregate filtering module required understanding how users navigate from a macro-level view of video data down to a specific problem code. Therefore, I needed to map out real-world use cases - for example, how a user would filter and isolate all video playback errors occurring across a live Major League Baseball broadcast.
2
Design for micro-level control:
Designing the session filtering module required understanding how users would isolate and analyze a continuous period of activity (a "session") for a single user, device, or system process. This allowed users to drill down from the aggregate view into the granular details.
3
Design an automated monitor and alert feature:
A latter piece of the project was the Anomaly Detection module, designed to enable rapid response times. However, we encountered a significant challenge with threshold sensitivity. The system would frequently trigger critical 'error' alerts for events that crossed a data threshold, but in reality, were only minor issues that mildly degraded video playback. This made it difficult for users to prioritize true emergencies.
To resolve the threshold sensitivity issue, I designed an alert configuration UI that empowered users to customize parameters for 'Info,' 'Warning,' and 'Critical' alerts. To streamline the user experience and prevent conflicting logic, I introduced a 'cascade up' mechanism. With this interaction, adjusting a baseline 'Info' threshold automatically scaled the 'Warning' and 'Critical' parameters alongside it, keeping the alert ecosystem balanced and drastically reducing manual setup time.
4
Not just dashboards: Custom dashboards:
The final phase of the project focused on building custom dashboards, allowing users to organize, combine, and monitor metrics relevant to their specific roles. To ensure a smooth transition and maintain user familiarity, I kept the existing advanced filtering method intact.
To ensure widget creation was intuitive, I implemented a drag-and-drop interface that allowed users to select a metric and easily link it to a filter. This streamlined the experience and eliminated unnecessary ambiguity.
5
Overall project efficiency:
The project was a 6 month contract with the goal of completing Aggregate Filtering and Anomaly Detection but Amazon works in small, yet nimble teams. This allowed me to go farther and complete 85% of the custom dashboard module before accepting my role at CBRE.