Project Summary
The Problem
Faculty grading was a major time burden, but research revealed that the true barrier wasn’t efficiency — it was trust. My initial exploration of two distinct grading flows (teacher-led vs. AI-guided) only added complexity, creating decision friction and amplifying faculty skepticism toward AI.
The Solution
I simplified grading into a single streamlined flow: default teacher-led grading, with the option to insert AI scores on demand or set AI as a default in settings. This gave cautious faculty a safe path, and curious faculty a fast path to experiment with AI.
The Impact
-
↓ Decreased: Average grading time 40%
-
↓ Decreased: Support tickets about grading by 30%
-
↑ Increased: Faculty adoption of AI workflows by 65% (post-pilot)
-
70% of faculty experimented with AI grading at least once
-
30% of those faculty later enabled AI Default Grading
-
Positioned UWorld as an AI innovator in education
Context & Problem
Faculty grading was a time-intensive bottleneck. Leadership’s request: “add AI to speed grading."
My early design explored two parallel grading flows (teacher-led vs. AI-guided). Through user interviews, faculty feedback was clear:
-
Faculty wanted slow, useful introductions to AI, not a wholesale replacement.
-
Teachers needed a default grading path that felt familiar and unobtrusive.
-
Curious faculty wanted the ability to experiment with AI, edit results, and validate them — but on their terms.
My Role & Responsibilities
-
Lead UX Designer: owned research, strategy, prototyping, and delivery.
-
Partnered with PM, engineers, and AI researchers.
-
Facilitated executive reviews, reframing the scope from “automating grading” to “building adoption through trust.”
Key Challenge:
How to introduce AI in a way that reduces workload without undermining trust.
Design Process & Iteration
Research & Insights
-
Faculty needed manual grading as the safe default.
-
Some were curious to try AI, but only if they could edit and validate scores easily.
-
Introducing two distinct flows created cognitive overload and slowed adoption.
-
Key Insight: Adoption depends on control + gradual introduction.
Iteration
We tested out prototypes with dual flows vs. single flow.
Simplified Flow
After gathering feedback one adaptive flow was created:
-
Teacher-led grading by default.
-
Faculty can insert AI scores at any time.
-
AI scores are fully editable before submission.
-
Optional settings toggle: “AI Grade by Default.”
Final Solution
The final grading flow created a dual-value system within a single experience:
-
Default path: Manual teacher grading → safe for those who distrust AI.
-
Optional path: Insert AI scores → edit/validate → submit.
-
Advanced path: Enable “AI Grade by Default” in settings → all students graded by AI upfront, with option to edit or override.
Outcome & Impact
-
↓ Decreased: Average grading time 40%
-
↓ Decreased: Support tickets about grading by 30%
-
↑ Increased: Faculty adoption of AI workflows by 65% (post-pilot)
-
70% of faculty experimented with AI grading at least once
-
30% of those faculty later enabled AI Default Grading
-
Positioned UWorld as an AI innovator in education
Reflection & Next Steps
Lesson Learned: AI adoption requires trust-first design — gradual exposure, optionality, and editability.
Next Steps
-
Expand human-in-the-loop model to rubrics, feedback, and reporting.
-
Add AI explainability features to further increase trust.
-
Continue measuring adoption curves (opt-in vs. default enablement).
This project demonstrated how design leadership bridges the gap between innovation and adoption — making AI workflows efficient, trustworthy, and human-centered.