Davide Molinelli
PhD in Computer Science · PostDoc Researcher at USI Lugano
About me
I am a postdoctoral researcher at the Università della Svizzera italiana (USI) and earned my Ph.D. in December 2025 with a dissertation that defines a neuro-symbolic approach combining AI models withstatic analysis to infer test oracles and automate the software verification process.
I enjoy integrating cutting-edge AI technologies into the entire software development lifecycle with the aim of automatizing routine tasks while improving software quality. My current research focuses on designingagent-based AI orchestrators that coordinate specialized agents to automatically generate entire test suites. This applied research aims to make the developed projects available for industrial use, thereby contributing to technological progress.
In addition to my primary research, I have a strong interest in UX/UI design, data analysis and visualization, as well as 3D web modeling
Skills
Research
My doctoral thesis — A Neuro-Symbolic Approach for Test Oracle Generation— centered on neuro-symbolic methods for the automated generation of axiomatic and concrete unit testing oracles, leveraging Java class documentation, method signatures, and test prefixes.
Currently, I am shifting the focus towards agent-based approaches in which an orchestrator delegates tasks to specialized agents for source code analysis, specification analysis, and generation of unit test cases and test suites.
Publications
- ICSEPredicting Failures in Smart Human-Centric EcoSystems2026
- ISSTATratto: A Neuro-Symbolic Approach to Deriving Axiomatic Test Oracles2025
- ASEDo LLMs Generate Useful Test Oracles? An Empirical Study with an Unbiased Dataset2025
- FSEHealth of smart ecosystems2021
- ICWEVoice-Based Virtual Assistants for User Interaction Modeling2021
Projects
Tracto
A neuro-symbolic framework for automatically generating concrete test oracles from Java method documentation, signatures, and test prefixes.
LLMs Oracle Generation
An empirical study evaluating the usefulness of LLM-generated test oracles using an unbiased benchmark dataset of Java methods. Presented at ASE 2025.
Tratto
A neuro-symbolic approach for deriving axiomatic test oracles, combining neural language models with symbolic constraint reasoning. Presented at ISSTA 2025.
Failure Predictions in Smart Human-Centric Ecosystems
Failure prediction and health monitoring for smart, human-centric software ecosystems. Presented at ICSE 2026.
Interests
Contact
Open to new opportunities.
Let's build something together.