Case study · June 2026

From a live walkthrough to working UI in under an hour.

Our team did not create STEPs. We worked like forward deployed engineers: learn Purdue Global's population and operating constraints, adapt the existing platform with its stakeholders, and leave a practical path forward.

Context

STEPs was an existing tutoring platform. Purdue Global needed it to fit working adults and other nontraditional learners, its faculty workflows, accessibility requirements, and its rules for how much help an AI tutor should provide.

The team

I worked with two fellow team members and Purdue Global stakeholders during the engagement. I led the live design sprint using Claude Design and helped turn stakeholder feedback into working UI. The post-event report has three authors, so this case study does not claim sole authorship of the broader engagement.

What shipped in the sprint

  • Lesson-linked resources and a get-unstuck assist for students.
  • Voice questions and read-aloud inside the STEPs chat.
  • An instructor view-as-student preview.
  • Faculty-defined stopping points for AI help.
  • Accessibility fixes to navigation and problem-opening controls.
  • Concepts for Canvas integration and resource surfacing.

Measured change

< 1 hourworking features against Purdue's expectation of months
1.5 → 4.8tool perception on a five-point pulse survey
1.8 → 4.5use-case fit on the same survey
3Purdue participants moved from skeptical to optimistic

Technical judgment

We recommended LTI 1.3 and LTI Advantage as the primary integration path, with vendor-specific LMS APIs behind adapter interfaces. We also recommended that the evaluation path avoid real Purdue student transcripts until de-identification, access control, and data-handling rules were approved.

Why this is forward deployed work

The job was not to arrive with a generic AI demo. It was to listen to a specific institution, make changes while stakeholders could react to them, and distinguish between what could ship quickly and what needed governance before real data entered the system.

Evidence boundary

The dates, survey figures, delivered features, and recommendations come from the substantially complete post-event report confirmed by one of its authors. The deployed prototype URL and private student or institutional data are not published.