TCC_MBA_Eng_Software: Research on the Integrated Application of Test-Driven Development and Large Language Models in Software Engineering

TCC_MBA_Eng_Software is an MBA software engineering thesis project published by VitorVA6 on GitHub (2026-06-06). Its core research focuses on the integrated application of Test-Driven Development (TDD) and Large Language Models (LLM), exploring new paradigms of AI-assisted software development. This research connects academic frontiers with industrial practices and has guiding significance for software engineering education and industry practices.

Original project link: https://github.com/VitorVA6/TCC_MBA_Eng_Software

TDD Practice and Challenges

TDD is a core practice of Extreme Programming, following the "Red-Green-Refactor" cycle: Red (write failing tests) → Green (write minimal code to pass) → Refactor (optimize structure). Its values include design-driven development, fast feedback, safety net, documentation role, and defect prevention, but there are challenges such as a steep learning curve, test maintenance overhead, slow initial speed, and difficulty in testing complex scenarios.

LLM Applications and Limitations

LLMs (e.g., GPT-4, CodeLlama) excel in code generation, completion, explanation, and test generation. Their advantages include rapid prototyping, reducing repetitive work, and assisting learning, but they have limitations like code errors, lack of business context, outdated training data, and hallucination issues.

Potential Integration Approaches

1. AI-assisted test generation: LLMs generate test cases based on functional descriptions to accelerate the Red phase
2. Test-driven code generation: Generate passing code using test cases as input
3. Refactoring assistance: LLMs suggest improvement plans and identify code smells
4. Test completion: Generate missing tests for legacy code

Research Questions

Core issues to explore include generated test quality, code-test matching degree, changes in development efficiency, code quality, and developer acceptance.

The project embodies the integration of theory and practice in MBA software engineering education, using empirical research methods: controlled experiments (traditional TDD vs. AI-assisted TDD), case studies (application in real projects), questionnaires (developers' opinions), and code analysis (quality metrics).

Knowledge contributions include best practices for AI-assisted development, new TDD models, human-machine collaboration paradigms, and suggestions for updating software engineering education.

Tool Trends

AI-native tools like GitHub Copilot and Amazon CodeWhisperer have integrated LLMs; the fusion of TDD and AI is a deepening of this trend.

Role Evolution

Developers' roles will shift to code reviewers and architects, spending more time on requirement understanding and design, and human-machine collaboration will become a core skill.

Quality Challenges

AI-generated code requires enhanced testing, updated review strategies, and evolved security scanning tools.

1. Progressive Adoption: Start with test generation for simple tool functions and gradually expand to complex scenarios
2. Critical Thinking: Review AI-generated tests and code, verify behavior, cover boundary cases, and ensure compliance with specifications
3. Feedback Loop: Record development speed, defect rate, review feedback, and satisfaction, then optimize strategies based on data.

Future Outlook

• Intelligent development environment: IDEs deeply integrate AI to automatically identify test paths and provide real-time test suggestions
• Test-first training: Specialized training for models that generate code compliant with tests
• Collaborative development: Developers define intentions, AI generates candidate implementations, and tests verify correctness

Summary

TDD and LLM are complementary; AI enhances traditional engineering practices (principles like test-first and fast feedback still apply). The project provides practical insights for the industry and represents a positive response of software engineering education to technological changes.