Boost Test Case Creation with AI: How a Multi‑Model Platform Cuts Effort by 80%

1. Background

AI capability explosion: With rapid AI advances, companies worldwide launch large AI models. Test case generation using these models is becoming a reality.

Company support for AI: KuJiaLe, a leading 3D design company, uses AI to improve efficiency and has an AI platform supporting various business lines, including a dedicated AI testing group.

Improving test case writing efficiency: Traditional manual test case creation is time‑consuming. AI can automatically generate initial test cases, which testers then review and refine, significantly reducing preparation time.

2. Platform Overview

2.1 Understanding the Platform

The platform is an AI‑large‑model‑based test case generation tool that integrates three AI engines, supports multiple input methods, online case editing, and two export formats.

Platform architecture diagram:

2.2 Three Input Methods

Direct generation: paste requirement text and click “Generate Cases”.

Image upload: upload a screenshot, adjust recognition results, then generate.

Free prompt: edit the platform‑provided prompt before generating.

2.3 Case Edit (Add/Delete/Modify)

The platform allows online editing of generated cases, including adding and deleting, as illustrated.

2.4 Export Options

Direct import into internal case management platform for review and test planning.

Export as .xmind file for local manipulation.

3. Case Generation Process

Because large‑model responses take tens of seconds, the system uses offline generation with front‑end polling to retrieve completed results.

Typical offline workflow: requirement storage → scheduled task fetches requirement → preprocessing → prompt assembly → GPT service generates cases → case parsing → retry on failure → task status update → case storage.

4. Tool Optimization

Early on, failure rates exceeded 50%. The team analyzed and mitigated issues.

4.1 Root Causes

Instability of a single GPT service.

Input length limits of GPT, causing failures for large texts.

Frontend technical issues leading to browser blocks.

4.2 Handling Service Instability

Added a retry mechanism (2 attempts), though impact was limited.

Introduced alternative models (Wenxin Yiyan and Minimax) as backups; if GPT fails, the backup engines generate cases.

4.3 Handling Length Limits

Backup engines alone produce lower quality. The solution combines Wenxin Yiyan for requirement understanding (product‑manager role) and GPT for case generation, leveraging both strengths.

4.4 Other Optimizations

Encrypt user input before sending to backend to prevent XSS attacks.

Provide free‑prompt feature allowing users to fine‑tune prompts for personalized generation.

5. Summary & Outlook

5.1 Achievements

The multi‑model test case generation tool offers full edit and export capabilities, has created over 300 generation tasks, produced more than 2000 cases, and achieved an 80%+ success rate, substantially improving tester efficiency.

5.2 Limitations

Lack of domain knowledge may lead to incomplete or inaccurate cases.

Limited handling of non‑functional requirements such as performance or security.

Complex systems may require deeper human expertise.

Absence of effective metrics to evaluate generated case quality.

5.3 Future Directions

Build a user center for self‑service task and result management.

Construct a knowledge base from existing cases to enhance AI generation precision.

Develop a case quality assessment mechanism that feeds back user edits to improve AI models.