How AI is Revolutionizing Mobile Test Case Creation with QAMate

Background

Test cases are essential for quality assurance across the entire development lifecycle, but creating and maintaining them—especially automated cases—has traditionally been costly and slow, limiting sustainable growth.

AI‑Driven Requirement‑Based Mind‑Map Cases

By leveraging Baidu's Wenxin large model, QAMate builds a standardized Prompt layer that converts textual requirements into mind‑map test cases. Over two months, the system generated and adopted more than 26,000 cases, effectively capturing and reusing business and testing knowledge.

Visual AI for Mobile UI Automation

QAMate replaces fragile XPath‑based element location with a YOLOv5‑based object detection model, OCR, and multi‑control layout algorithms. Users simply tap their phones to record actions, reducing the per‑step authoring time from 40 seconds to 5 seconds and achieving an average execution stability above 90%.

80% of automation scenarios involve simple single‑control detection and click actions, solved by a generic visual locator.

The remaining 20% require multi‑control coordination, addressed by a combined visual and DOM algorithm.

Traffic‑Driven API Test Case Generation

Using real‑time traffic captured via eBPF, XSTP GoReplay, or file uploads, QAMate automatically generates API test cases. This approach saves up to 70% of manual effort, with generated case coverage increasing from 14.8% to 46.7% and line coverage from 9.9% to 34.7%.

Four traffic sampling strategies: lightweight, priority, normal, and high‑coverage.

Multiple fault‑injection strategies: enumeration, boundary, required‑field, type errors, etc.

Prompt Layer and Knowledge Base

The open Prompt layer enables each business unit to define input‑output rules, optimizing large‑model performance for specific scenarios. An external knowledge base stores historical case features, allowing personalized test‑case generation that directly satisfies business requirements without manual adjustment.

End‑to‑End Workflow Integration

Generated cases are fed back into the product loop, automatically updating the visual model based on execution metrics, forming a data flywheel that continuously improves AI capabilities.

Key Practices

Low‑cost traffic ingestion via eBPF, requiring only a BNS identifier.

Support for long‑connection, SSE, and chunked interfaces.

Configurable sampling and case‑generation strategies to exceed human‑maintained coverage.

LUI + LLM enables complex parameter and assertion modifications while preserving test reliability.

Future Outlook

Beyond case generation, AI can assist in execution, transformation, updating, and retirement of test cases, creating a full lifecycle that continuously enhances development‑testing efficiency.