The challenge was not simply a lack of information. Complex repair combined proprietary product logic, overlapping symptoms, and fragmented support resources. Traditional troubleshooting trees were often too rigid, while generic AI chat risked sounding helpful without being reliable enough for hands-on repair.

Research therefore focused on where novice users lost confidence, how repair knowledge was distributed, and which repair scenario could meaningfully test a guided AI workflow.

Survey findings and expert input helped identify the boost-button failure as a suitable scenario for testing diagnosis, troubleshooting, and physical repair within one workflow.

A conventional chatbot could answer repair questions, but repair could not depend on plausible guesses. The product challenge was therefore to turn conversational AI into a guided troubleshooting workflow that reduced uncertainty, preserved context, and helped users act safely step by step.

Early concepts explored different ways of combining physical product context with conversational support. Across the iterations, the same issue emerged: users needed one coherent workspace that could preserve progress, surface relevant evidence, and support dialogue without forcing them to mentally reconstruct the repair state.

This shifted the design away from a generic chat interface toward a contextual repair hub.

Iteration 01 — Spatial Mapping

Centered the product itself as the navigation surface. It improved physical orientation but provided weak support for open-ended troubleshooting.

Iteration 02 — Split View

Introduced persistent progress tracking and conversational assistance, but still separated guidance across competing visual surfaces.

Final Direction — Contextual Hub

Combined progress tracking, contextual evidence, and AI dialogue within one workspace so users could understand both where they were and what to do next.

The final experience transformed the assistant from a question-answer interface into a guided repair workflow. Each stage reduced uncertainty while preserving enough context for users to understand what the system was doing and why.

To test whether the assistant could support not only diagnosis but real repair behaviour, I built a functional prototype that connected conversational guidance to a physical boost-button replacement task.

Twenty participants completed a scenario spanning issue diagnosis, troubleshooting, and hands-on repair, allowing the concept to be evaluated as a complete human–AI–physical interaction rather than a screen-only interface.

The evaluation indicated that participants could move through a technically complex repair scenario while maintaining high usability and manageable perceived workload.

The project continued beyond the graduation study as a demonstrator of how grounded conversational AI could support complex service and repair contexts.

Its later presentation across academic and industry settings extended the work from a single prototype into a broader discussion around trustworthy AI, repair, and human-centred service design.

The project reinforced that AI assistance becomes most valuable in complex tasks when it helps users understand uncertainty and move forward—not when it simply produces more confident answers.

Designing for repair therefore required treating grounding, transparency, progression, and physical action as parts of the interaction model rather than backend technical concerns.