AI‑Driven Automated Testing System: Architecture, Workflow, and Benefits

The development process traditionally involves separate development and self‑testing phases, where manual interface testing after code changes is time‑consuming, error‑prone, and often misses impact assessments. To address these pain points, an AI‑automated testing system was introduced.

Long self‑testing cycles reduce development efficiency.

Incomplete impact assessment can overlook potential issues.

Manual testing is error‑prone and hard to guarantee quality.

By leveraging AI, the system automates test case invocation, significantly saving time and allowing developers to focus on coding.

Problem Solved

Efficiency Effect

Improved self‑testing efficiency

Previously, developers manually executed many test cases, which was labor‑intensive and error‑prone. With AI automation, a single click runs all interface invokes, drastically reducing time and effort.

Enhanced impact‑area assessment

Earlier, incomplete impact analysis left some affected interfaces untested, creating hidden risks. The AI system precisely identifies all impacted interfaces and executes corresponding test cases, eliminating omissions.

The system follows a modular design consisting of four core components:

Code Analysis Module – handles code release triggers, application info retrieval, code cloning & analysis, and diff analysis.

AI Test Engine Module – performs AI‑based code analysis, constructs parameters using the "Neighbor Pattern", reviews test results, and tags them intelligently.

Automation Execution Module – supports multi‑environment invoke and result comparison.

Report Generation Module – creates and pushes test reports.

Code Analysis Module

3.1.1 Code Release Trigger

Developers publish code in the development environment.

The system automatically monitors code changes.

Triggers the automated testing workflow.

3.1.2 Application Information Retrieval

Retrieves via development platform API: application name, branch, Git repository URL, and commit information.

3.1.3 Code Cloning and Analysis

The system clones the remote Git repository into a container, switches to the specified branch, and prepares the test environment.

3.1.3.1 Code Cloning

Automatically clones remote code into a container.

Switches to the target branch.

Prepares the testing environment.

3.1.3.2 Code Analysis

Uses javaparser to build an AST, identifying classes, methods, and annotation dependency networks.

Parses code structure and dependency relationships.

Constructs a code call‑graph.

3.1.4 Code Diff Analysis

Creates a temporary test branch.

Uses git diff to analyze changes.

Precisely locates modified methods.

Analyzes the scope of impact.

AI Test Engine and Automation Execution Modules

The AI‑driven workflow includes method analysis, interface recognition, multi‑environment comparison, and intelligent parameter construction.

3.2.1 Method Analysis & Interface Identification

Static analysis locates related Dubbo service entry points, then an AI model performs deep semantic analysis to assess executability.

3.2.2 Multi‑Environment Comparison

Dubbo calls are executed in both base and development environments; any result differences trigger an AI‑powered comparison engine to determine true business logic changes, reducing false positives.

3.2.3 Intelligent Parameter Construction – Neighbor Pattern

When direct input parameters are unavailable, the system employs a "Neighbor Pattern" that extracts parameters from semantically similar methods within the same class, feeds them to the AI model, and generates context‑aware test inputs.

Context analysis: find similar methods.

Parameter extraction from those methods.

AI‑driven mapping and generation of realistic parameters.

Advantages include business relevance, scenario fit, data accuracy, and automatic handling of parameter dependencies.

Test Reporting

The AI system records test execution, provides overview dashboards, and highlights discovered bugs.

Results

Automatic execution after code release in development or base environments, with reports pushed to group chats.

Thousands of test runs executed to date.

Hundreds of effective bugs discovered.

Accumulated hundreds of thousands of test case parameters for future reuse.

Future Plans

AI‑driven front‑end page simulation using browser‑use and prompts.

AI‑generated unit tests that compare pre‑ and post‑change code, push tests to a dedicated branch, and execute them to cover most code‑change scenarios.