2024
APITestGenie
An LLM-powered tool that writes executable API tests from business requirements, cutting developer testing time by 4x.
- generative-ai
- llm
- api-testing
- research
Problem
API tests are written by hand: developers read the business requirements and the API specification, then turn them into test scripts. It is slow, repetitive work, and it competes with building the product.
Approach
I built APITestGenie as an AI researcher at Critical TechWorks (BMW Group), as my master thesis (January to July 2024). The tool uses large language models to generate executable API test scripts from two inputs: the business requirements and the API specification.
We evaluated it on 10 real-world APIs and asked industry specialists for feedback.
Outcome
- Valid test scripts in 57% of attempts, rising to 80% with three generation attempts per task.
- Developer testing time reduced by 4x.
- Industry specialists expressed strong interest in adopting it.
- Positioned as a productivity assistant for testers, not a replacement: generated tests should be reviewed by a person before they go into a CI/CD pipeline.
The work became a peer-reviewed paper with Bruno Lima and João Pascoal Faria, which I presented at ICSE 2026 (IEEE/ACM International Conference on Software Engineering) in Rio de Janeiro.