About me

  • Bachelor’s and Master’s Degree
    in Computer Engineering
  • Research Internship
    in Generative AI
  • Ph.D. Degree
    in Computer Science


  • Postdoc
    in Software Engineering
  • Great Minds Fellowship
    in Software Security


Short Bio:
Luca Di Grazia holds a Great Minds Fellowship at the University of St. Gallen in Switzerland, hosted by Prof. Dr. Guido Salvaneschi on security of AI-generated code. Previously, he was a Postdoctoral Researcher at USI Lugano, focusing on GenAI for software testing. In this role, he supervised PhD students and designed real-world testing scenarios in the group of Prof. Dr. Mauro Pezzè. He earned his Ph.D. in Computer Science (summa cum laude) from the University of Stuttgart, under the supervision of Prof. Dr. Michael Pradel. His dissertation, "Supporting Software Evolution through Search and Predictions", explored innovative approaches to software maintenance and development. During a Research Internship at Uber in Amsterdam, he developed a GenAI-based bug-fixing approach for Java linter errors. His project won an internal Uber competition against 103 teams, and the team presented the results directly to the Uber CEO.

Three Major Achievements:
  • 2025: Great Minds Fellowship funded by the University of St. Gallen with 189,747 CHF (∼$230,000) for the research project "Secure Software in the Age of Massively GenAI-Generated Code". Principal investigator.
  • 2023: Won GenAI Uber competition with his internship project on using generative AI for fixing bugs to boost developer productivity, beating 103 teams, and presenting as winners to the Uber's CEO and ELT.
  • 2022: Won the ACM SIGSOFT Distinguished Paper Award at ESEC/FSE.

Research Interests:
Generative AI to automatically generate test oracles (ASE 2025), fix Python type annotations (ICSE 2024), Java liter errors (Uber research internship), Software Evolution and Mining Repositories (ESEC/FSE 2022, Distinguished Paper Award), Code Change Retrieval (IEEE TSE paper, ICSE 2022 SRC 2nd winner), Code Search (ACM CSUR paper), and more. Additionally, He has supervised a total of seven students, including projects on "Towards Automatically Repairing Errors in Python", "Testing FAISS indexing on DiffSearch", on "Improving the Recall of Searching for Code Changes", and more.

Contact: Torstrasse 25, 9000 St. Gallen, Switzerland.
Institutional: luca.digrazia@unisg.ch Private: work@lucadigrazia.com
Extra Info:
References are available on request. All opinions expressed in this website are his own and not of his employer. No information is gathered from you by this website.

Latest News

10/2025: ICSE 2026! Our paper: "E-Test: E'er-Improving Test Suites” has been accepted at ICSE 2026 in Brazil.

10/2025: I presented our ASE 2025 paper at University College London, invited by Prof. Federica Sarro.

10/2025: I taught a course (1 ECTS) on AI for Data Analysis at Bern University of Applied Sciences.

08/2025: Starting a Great Minds Fellowship (∼$230,000) at University of St. Gallen in Switzerland, hosted by Prof. Dr. Guido Salvaneschi.

08/2025: ASE 2025! Our paper: "Do LLMs Generate Useful Concrete Test Oracles? An Empirical Study with an Unbiased Dataset” has been accepted at ASE 2025 in South Korea.

06/2025: Ketai Qiu (PhD I co-supervise) won the ACM Student Research Competition at FSE 2025!

01/2025: Invited to serve as a co-chair for the Tool Demonstration Track of Internetware 2025.

12/2024: Invited to serve as a reviewer for the International Conference of Software Engineering (ICSE 2026).

07/2024: Our paper about: “Devs in 2030” received significant media attention on The Register and Swiss IT Magazine .

04/2024: Attended a fantastic Dagstuhl Seminar on Code Search!

02/2024: Successfully defended my PhD (summa cum laude) and starting a new position at USI advised by Prof. Mauro Pezzè in Switzerland.

12/2023: ICSE 2024! Our paper: “PyTy: Repairing Static Type Errors in Python” has been accepted for the ICSE 2024 conference in Portugal.

06/2023: Won a Uber competition on Generative AI for developer productivity among 103 teams worldwide.

05/2023: Generative AI at Uber! I am joining Uber for a research internship in Amsterdam.

03/2023: I was invited by JetBrains in their Munich office to discuss our paper “DiffSearch: A Scalable and Precise Search Engine for Code Changes”.

12/2022: Distinguished Paper Award! Our paper “The Evolution of Type Annotations in Python: An Empirical Study” received an ACM SIGSOFT Distinguished Paper Award at ESEC/FSE 2022 in Singapore.

Selected Peer-reviewed Publications

E-Test: E'er-Improving Test Suites

ICSE 2026

E-Test leverages LLMs to pinpoint scenarios that the current test suite does not adequately cover, and augments the suite with test cases that execute these scenarios. Our evaluation on a dataset of 1,975 scenarios demonstrates that E-Test retrieves not-yet-tested execution scenarios significantly better than state-of-the-art approaches. While existing regression testing and field testing approaches for this task achieve a maximum F1-score of 0.34, and vanilla LLMs achieve a maximum F1-score of 0.39, E-Test reaches 0.54.
Do LLMs Generate Useful Concrete Test Oracles? An Empirical Study with an Unbiased Dataset

Do LLMs Generate Useful Concrete Test Oracles? An Empirical Study with an Unbiased Dataset

ASE 2025

This paper presents an empirical study of the effectiveness of LLMs in generating test oracles. Our experiments use 13,866 test oracles, from 135 Java projects, that were created after the LLM training cut-off dates. Thus, our dataset is unbiased.

Do LLMs Generate Useful Concrete Test Oracles? An Empirical Study with an Unbiased Dataset

From Today’s Code to Tomorrow’s Symphony: The AI Transformation of Developer’s Routine by 2030

ACM TOSEM 2025
News press articles from The Register and Swiss IT Magazine

We envision HyperAssistant, an augmented AI tool that offers comprehensive support to 2030 developers, addressing current limitations in mental health support, fault detection, code optimization, team interaction, and skill development.

Finally, many developers seem to not regularly check their code for statically detectable type errors, or if they do, commit the code despite such errors.

PyTy: Repairing Static Type Errors in Python

ICSE 2024

PyTy is a novel automated technique aimed at fixing type errors in Python. It was developed based on a study and employs a dataset named PyTyDefects, containing over 2,700 Python type errors fixes. The paper highlights the use of cross-lingual transfer learning to enhance PyTy's effectiveness, even with a small dataset. This involves adapting an existing program repair model for PyTy's use. PyTy proved highly effective, successfully resolving 85.4% of type errors in the evaluation. Additionally, its real-world applicability is shown by the high acceptance rate of GitHub pull requests using PyTy's fixes.
Finally, many developers seem to not regularly check their code for statically detectable type errors, or if they do, commit the code despite such errors.

The Fault in Our Stars: Designing Reproducible Large-scale Code Analysis Experiments

ECOOP 2024

Large-scale software repositories offer insights into software development but pose scaling challenges. This paper proposes a standardized methodology for sampling and studying repositories, emphasizing clear population definitions, reproducibility, and avoiding unreliable metrics like project popularity.

We advocate for a methodology that clearly lays out what the population of interest is, how to sample it, and that fosters reproducibility.

Code Search: A Survey of Techniques for Finding Code

ACM Computing Surveys (CSUR), 2023

This article provides a comprehensive overview of 30 years of research on code search. Given the huge amounts of existing code, searching for specific code examples is a common activity during software development. We discuss what kinds of queries code search engines support, and give an overview of the main components used to retrieve suitable code examples. In particular, the article discusses techniques to pre-process and expand queries, approaches toward indexing and retrieving code, and ways of pruning and ranking search results.
the article discusses techniques to pre-process and expand queries, approaches toward indexing and retrieving code, and ways of pruning and ranking search results.

The Evolution of Type Annotations in Python:
An Empirical Study

ESEC/FSE 2022
ACM SIGSOFT Distinguished Paper Award

We extract 1.4 million type annotation changes from 9,655 Python repositories. Our results show that type annotations are clearly gaining traction, yet the large majority of code elements that could be annotated currently remains unannotated. We see a huge potential for techniques that automate the process of adding types into an existing code base, such as neural type prediction models. Finally, many developers seem to not regularly check their code for statically detectable type errors, or if they do, commit the code despite such errors.
Finally, many developers seem to not regularly check their code for statically detectable type errors, or if they do, commit the code despite such errors.

DiffSearch: A Scalable and Precise Search Engine for Code Changes

IEEE Transactions on Software Engineering (TSE), 2022
2nd prize winner at ICSE Student Research Competition 2022
and Journal-first track at ASE 2023

We present a scalable and precise search engine for code changes. Given a query, the approach retrieves within seconds relevant examples from million code changes. Our query language extends the underlying programming language, providing an intuitive way of formulating queries to search for code changes. DiffSearch guarantees that every returned search result fits the query.
We present a scalable and precise search engine for code changes.

Imprint © MIT License.