Curry Reduces Test Case Creation Effort by 36.7% and Automates Defect Registration with QA AI Hub

• Built an AI-based platform to solve the problem where more than half of total QA time was spent on TC creation
• Real-world measurements across 137 tasks showed that AI replaces 70% of TC work, increasing the number of tasks handled per person by 59% per month
• When correct TCs and domain policy documents are available, AI generation quality rises to 95 points, indicating that reference assets determine quality
• Lack of test data accounts for 71% of execution bottlenecks, making automated data generation and converting personal know-how into team assets the next challenge

Key Performance Indicators

• In the commerce domain, preparation time per task decreased by 35%, and preparation effort per 1,000 TCs decreased by 36.7%
• AI execution reliability verification results showed that the false PASS rate remained low at 0.64%
• The logistics domain requires more refinement effort than twice that of commerce due to integration complexity, necessitating differentiated application by domain

Technical Architecture

• Adopted a distributed structure consisting of a Next.js 15 server and Node.js workers on individual QA PCs
• Used Claude Code, Playwright, and Mobile MCP to manipulate screens based on natural language TCs
• Applied skill compliance verification and Claude Code hooks as a dual gate to block the leakage of substandard deliverables