AI-Assisted JUnit Testing

Portfolio case study based on a Java/JPA university assignment: generate unit tests with one AI prompt, review the output, and organize the result into a runnable Maven test suite.

Project Idea

The original assignment implemented a small JPA model for a TOVAR table and three static methods:

• Application.novyTovar(String nazov, Double cena)
• Application.tovarBezCeny()
• Application.odstranTovarBezCeny()

The AI task was to generate JUnit 4 tests that grade the solution from 0 to 10 points and verify persistence behavior, database constraints, duplicate handling, null prices, JPQL queries, and delete operations.

Repository Structure

Path	Purpose
src/main/java/vsa/	Sanitized Java/JPA assignment solution
src/test/java/vsa/	Runnable Codex-generated JUnit 4 test suite
ai-generated/gemini/	Gemini-generated Java test outputs kept for comparison
ai-generated/chatgpt/	ChatGPT-generated Java test outputs kept for comparison
prompts/codex-prompt.md	The one-prompt test-generation request
docs/assignment-summary.md	Short summary of the original assignment
docs/test-results.md	Captured test-result summary
docs/comparison.md	Notes comparing the generated approaches

Run The Tests

Requirements:

• Java 21
• Maven 3.9+

mvn test

The Maven project uses an in-memory H2 database in PostgreSQL compatibility mode so the tests can run without a local PostgreSQL server.

Current verification:

• mvn test passes locally with Java 21 and Maven 3.9.9.
• The runnable Codex suite prints BODY = 10 for both ApplicationTest and ApplicationTestAlt.
• EclipseLink may print expected warnings during constraint checks because the tests intentionally try invalid NULL and duplicate inserts.

What The Tests Check

• A new product is persisted and receives an ID.
• Duplicate product names return null instead of throwing.
• NAZOV is unique and not nullable.
• CENA can be null.
• tovarBezCeny() returns all products without a price.
• odstranTovarBezCeny() removes only products without a price.

Portfolio Value

This project shows more than "AI wrote tests." It demonstrates:

• turning an assignment specification into testable requirements;
• using AI as a test-generation assistant;
• reviewing generated tests for coverage and correctness;
• separating runnable tests from comparison artifacts;
• documenting results and tradeoffs.

Raw submission archives, saved browser pages, account-related browser assets, and .docx logs are intentionally excluded.