Safety Driving Education System
Personalizing driver education with AI to halve traffic fatalities by 2030.
- Role
- GenAI Engineer & UI/UX Design
- When
- Oct 2025 – Feb 2026
- Context
- Honda Motor Co.
- Built with
- Figma, FastAPI, LangChain, Next.js, Azure OpenAI
Overview
A comprehensive AI-powered driver education platform that personalizes safety training through psychometric profiling and adaptive AI instruction. Designed and built an end-to-end experience spanning 19 screens — from registration and DSQ personality assessment through AI-generated curriculum, interactive lessons with a conversational AI instructor, to e-certification.
The problem
Traditional driving education uses a one-size-fits-all curriculum that fails to address individual risk profiles. Aggressive drivers receive the same training as cautious ones. Meanwhile, instructor shortages — especially in low- and middle-income countries — limit access to quality safety education.
Decisions and status
Legacy prototype vs. redesign
What the audit found, and what the lesson page does now.
Legacy prototype
- Lesson audio auto-plays with no pause
- Progress bars render invisibly
- White text on white backgrounds
- Bilingual text breaks across screens
- No error recovery in the AI chat
Redesign
- Pause control and visible progress
- A source citation on every answer
- Rulebook images shown inline
- Objective and topic completion kept separate
- Supportive, non-punitive feedback
Decision card
The call that shaped the project.
Where it stands
Delivered work and predictions, kept apart.
- In productionDeployed for foreign-license-conversion training.
- PredictedAbout 15% higher safety awareness, from a Phase 1 simulation. Largest for aggressive and risk-prone profiles.
- ~15%Predicted safety awareness
- 19Screens designed
- 18QDSQ assessment
- 9Personality traits profiled
Research and discovery
- Driving Style Questionnaire (DSQ) — psychometric framework for risk profiling across 9 traits
- Existing driving school curriculum analysis — gap identification
- UX audit of legacy system — 8+ critical usability issues documented
- Stakeholder alignment — Honda Safety Department, management, and learner needs
Key insight
The biggest UX challenge wasn't the AI — it was trust. Learners needed to trust an AI instructor enough to engage honestly with their driving weaknesses.
Design process
UX audit — documented critical usability failures in the legacy system
19-screen user journey: registration → DSQ → curriculum → AI lessons → certification
DSQ results page with radar chart visualizing 9 personality traits
AI instructor chat interface with source citations grounding every answer
Comprehension checks with supportive, non-punitive feedback
Celebration micro-animations for lesson and course completion
Figma workflow
User journey mapping and information architecture — sketching the end-to-end flow from registration through DSQ assessment, AI instruction, and e-certification before committing to any screen design.
Wireframing and prototype exploration — rapid low-fidelity iteration on layout structures, navigation patterns, and the DSQ radar chart visualization before moving to high-fidelity.
High-fidelity screens and component design — 19 production-ready screens with a bilingual EN/JP component library (React 19 + Tailwind CSS v4 + Radix UI).
Dev handoff — as both designer and engineer on this project, the Figma MCP server let Claude Code read frames directly and generate React components without manual export. Result: design and production stay in step across all 19 screens. Target: a 3–4× faster design-to-component cycle, not yet measured.
The pivot
The original lesson page auto-played audio with no pause control, had invisible progress bars, and showed white text on white backgrounds. The UX audit revealed 8+ critical issues. The entire experience was redesigned from scratch rather than patching the existing system.
Results
- Deployed to production for foreign-license-conversion training
- Well received by the Honda Safety Department and management; the redesigned UX was singled out as the strongest delivered output
- Predicted ~15% improvement in safety awareness from a Phase 1 simulation (LLM-as-judge across driver personas), the evidence base for the product
- Largest predicted improvements for aggressive and risk-prone driver profiles
- Co-authored a conference paper
Reflection
Designing AI-powered experiences is fundamentally about designing trust. Celebration, encouragement, and clarity aren't nice-to-haves — they're essential for behavior change.
- AI Product Design
- UX Audit
- Psychometric UX
- Safety-Critical