Introduction

A leading public transportation operator in Japan aimed to improve how it evaluates staff professionalism, particularly customer service attitude and appearance. Historically, these aspects were assessed subjectively, making it difficult to track progress or recognize high performers.

The goal of this PoC project was to design an AI-driven system that could quantify previously subjective criteria, allowing for clearer feedback, fairer evaluations, and more consistent service quality.

Team size

5 members

Industry

Transportation

Technology

AWS S3, AWS EC2, AWS CloudWatch, Qwen2.5-VL, GLM, Python, Alembic, SQLAlchemy, ReactJS, Vite, WebSocket, Jenkins


Highlights

  Value Delivered

  • Quantified appearance scores: Provided clear and actionable feedback
  • Motivated staff through transparent and fair evaluations
  • Improved service quality by encouraging a professional public-facing image
  • Protected privacy using open-source AI models and in-house infrastructure
  • Scalable foundation for future assessment areas

Challenges

  • Staff evaluations were based on gut feeling and personal judgment
  • No measurable criteria for improvement or recognition
  • High-performing employees lacked visibility and motivation
  • Inconsistent assessments across departments and locations

Solutions

  • AI Image Analysis: All images are analyzed privately using a self-hosted AI model within a secure cloud environment, no third-party services involved.
  • Role-Based Access Control:
    • Admin: Manages users, prompts, and evaluation settings
    • Manager: Captures and uploads staff images
    • Staff: Views personal evaluation results
  • Prompt Management: Allows fine-tuning of evaluation logic without changing the codebase
  • Cloud-Native Architecture: Deployed entirely on AWS for scalability, security, and performance
  • Data Privacy: No third-party AI services used, all models are hosted privately