SQL Scrolls, a learning game created by FHNW, uses multiple technologies: a JavaScript frontend, a Node.js backend, Python/FastAPI for recommendations, and MongoDB. Over time, several features, including a recommendation service, a leaderboard, and an AI teaching assistant, have been added by various developers. Each new feature requires manual verification, which is time-consuming and error-prone. Additionally, the AI Teaching Assistant provides a new feature whose practical value has yet to be assessed by users.

We designed a layered testing strategy, including unit, integration, and end-to-end tests: Jest for backend and frontend logic and API routes, Pytest for the recommendation service, and Playwright for essential user journeys. All tests are integrated into a CI pipeline that runs them automatically, generates a coverage report and enforces a 95% coverage gate. Additionally, we created a test environment with Infrastructure as Code, integrated and deployed the AI assistant and conducted a crossover user study with current and former students.

SQL Scrolls now includes a comprehensive and maintainable test suite that automatically verifies API contracts, UI regressions, and full-stack behaviour. The CI pipeline prevents merges when tests fail or coverage is insufficient, and it publishes reports and traces, eliminating the need to manually reproduce errors. This streamlines quality assurance efforts and reduces the risk of errors and bugs. Additionally, it supports future developments and simplifies onboarding new developers, making the process more consistent and straightforward. In parallel, the AI teaching assistant was evaluated in a classroom-ready deployment. Survey results were strongly positive (e.g., "helps solve tasks" mean rating of 4.6/5), and participants completed more tasks with AI available (avg. 4.40 vs 1.87 tasks; mean gain +2.53). These findings support an evidence-based recommendation for rollout and identify areas for improvement before production use.