Infosys:
AI-Augmented
Learning Design.
Designing GenAI rubric engines and recommendation systems for enterprise-scale educational platforms at Imagine Learning — a K-12 EdTech platform serving students across the US.
Full case study available under NDA. Contact me directly for a confidential walkthrough.
Key Outcomes
Scope & NDA
This work is covered under a Non-Disclosure Agreement with Infosys and Imagine Learning. Key outcomes and high-level contributions are publicly shareable — design artifacts, full case study documentation, and detailed process notes are available for sharing in a confidential context. Contact me directly to schedule a walkthrough.
01 — Outcomes
What We Shipped
Reduction in assessment time
AI rubric engine vs. manual baseline (time-on-task study, n=28 teachers, pre/post). Specific data available under NDA.
Product teams served
Unified design system components adopted across Imagine Learning's product portfolio.
AI tools shipped to production
Including GenAI rubric generators, adaptive recommendation engine UX, and feedback tools.
02 — Contributions
What I Did
- 01
Designed UX for GenAI-powered rubric generation engines — reducing manual evaluation from hours to minutes
- 02
Led design system expansion for Imagine Learning — adding AI-specific interaction patterns and states
- 03
Created end-to-end flows for adaptive content recommendation systems
- 04
Ran design sprints with cross-functional teams (ML engineers, curriculum designers, product managers)
- 05
Established accessibility standards (WCAG 2.2) across all AI-facing features
- 06
Delivered high-fidelity prototypes for executive stakeholder reviews
02.5 — Design Challenges
The Hard Problems
Three design tensions that defined the engagement — shareable without violating NDA.
AI Confidence ≠ User Trust
The rubric engine outputs confidence scores, but showing raw percentages caused teachers to either over-trust (80% feels like a guarantee) or dismiss (60% feels unreliable). Designed a 3-tier signal system (Verified / Suggested / Uncertain) that anchored decisions in pedagogical context, not probability.
Outcome
Teacher acceptance rate increased from 41% to 74% in A/B testing.
GenAI Latency UX
Rubric generation takes 3-8 seconds — far longer than users expect from 'AI'. A generic spinner caused drop-off. Designed a progressive reveal: skeleton rubric structure appears immediately, then cells populate sequentially, creating a perception of real-time generation.
Outcome
Drop-off during generation reduced by 55%.
One system, five teams
Five product teams had divergent component needs — elementary literacy, middle-school math, special ed, teacher dashboards, admin tools. Designed a token layer with 4 semantic contexts (learner, educator, admin, assessment) that let the same components adapt across surfaces.
Outcome
Unified system with zero hard forks across 5 products.
03 — Work Samples
The Product We Built
High-fidelity screens from the AI tooling suite. Full documentation available under NDA.
AI Rubric Generator — generates assessment criteria from a learning objective in seconds, with contextual AI suggestions.
Adaptive Recommendation Engine — surfaces personalised content per student based on ML-identified skill gaps.
04 — Design Iterations
How the AI Confidence Signal Evolved
3 rounds of internal testing with teachers before the confidence signal shipped. Each round invalidated a prior assumption.
Teachers over-trusted 80%+ scores as guarantees; dismissed 60%- scores entirely. Binary thinking, not probabilistic.
Drop percentages. Switch to qualitative signals.
Green triggered rubber-stamping — teachers stopped reading the rubric text. 'Green means approve' was too automatic.
Remove green. Reframe as signal, not verdict.
Teachers engaged differently at each tier — Verified rubrics were approved 2× faster; Uncertain triggered review. Acceptance rate: 41% → 74%.
Ship this version. Anchored in pedagogical action, not probability.
Want to see the full case study?
I'm happy to walk you through the work, process, and outcomes in a confidential setting.