About
I'm Matteo Gabburo, CTO of Aithlas, where I lead the design and deployment of AI-driven solutions for industrial clients. I hold a PhD in Information and Communication Technology from the University of Trento, where my research focused on loss and reward functions for Generative Question Answering systems under the supervision of Prof. Alessandro Moschitti.
My research spans training generative QA models with weak supervision techniques, developing automatic evaluation systems for QA, and designing efficient pre-training tasks for Transformer models.
NLP
Question Answering
Generative QA
Automatic Evaluation
Transformers
Language Models
Retrieval-Augmented Generation
Experience
Chief Technology Officer
Jun 2025 – Present
Vicenza, Italy · Remote
Leading the design and deployment of AI-driven solutions for industrial clients across B2B and B2C contexts. Responsible for end-to-end technical strategy, model development, and production integration of machine learning systems.
Applied Scientist
Amazon
Oct 2024 – Mar 2025 · 6 mos
Berlin, Germany
Context Selection Techniques for Efficient Retrieval-Augmented Generation. Proposed a lightweight LLM-based content selection method achieving up to 98% reduction in input length and 5.2% accuracy improvement across five QA benchmarks.
Applied Scientist
Amazon
Jul 2023 – Oct 2023 · 4 mos
Santa Barbara, CA, USA
Question Decomposition for Retrieval-Augmented Answer Generation. Analyzed benefits and limitations of question decomposition strategies in large-scale RAG systems.
Applied Scientist
Amazon
Feb 2022 – Feb 2023 · 1 yr 1 mo
Los Angeles, CA, USA
Automatic Evaluation Metrics for Seq2Seq Answer Generation Models. Developed novel automatic evaluation methodologies for QA systems and enhanced training strategies for generative QA models.
Applied Scientist
Amazon
Apr 2021 – Oct 2021 · 7 mos
Los Angeles, CA, USA · Remote
Knowledge Transfer from Answer Selection to Text-to-Text Models. Designed novel knowledge distillation techniques to transfer supervision from AS2 models to seq2seq architectures, published at EMNLP 2022.
Early Stage Researcher
University of Trento
Mar 2019 – Sep 2020 · 1 yr 7 mos
Trento, Italy
Transformer-Based Models for Program Completion. Leveraged Transformer architectures to design and evaluate multilingual models for source code, in collaboration with Orange.
Education
November 2020 – March 2025
PhD in Information and Communication Technology
University of Trento
Trento, Italy
Advisor: Alessandro Moschitti. Thesis: “Loss and Reward Functions for Generative Question Answering Systems”
September 2016 – March 2019
MSc in Computer Science
University of Trento
Trento, Italy
Advisor: Andrea Passerini. Thesis: “Learning Activation Functions for Type Extension Trees”
September 2012 – March 2016
BSc in Computer Science
University of Trento
Trento, Italy
September 2007 – June 2012
Secondary School Diploma in IT
ITIS Enrico Fermi
Bassano del Grappa, Italy
Publications
Loss and Reward Functions for Generative Question Answering Systems
PhD Dissertation, University of Trento, 2025
SQUARE: Automatic Question Answering Evaluation using Multiple Positive and Negative References
IJCNLP-AACL 2023, Main Conference
Learning Answer Generation using Supervision from Automatic Question Answering Evaluators
ACL 2023, Main Conference
Evaluating Retrieval System for Language Model Processing
US Patent 12,579,174, 2025
Peer Reviewing
Conferences
2025 ARR 2025
2024 ACL, EACL
2023 IJCNLP-AACL, EMNLP, ECMLPKDD, ACL
2022 WSDM, ICML, EMNLP, NAACL, KDD
2021 EMNLP, ACL-IJCNLP
2020 CLIC-IT
Journals
Journal of Artificial Intelligence Research (JAIR) · ACM Computing Surveys (CSUR) · Journal on Data Semantics (JoDS) · The Computer Journal (COMJ)
Skills
Programming Languages
- Proficient: Python, Java, C, C#, SQL, NoSQL
- Familiar: C++, JavaScript, OCaml, x86 Assembly
ML Frameworks
- Proficient: PyTorch, PyTorch Lightning
- Familiar: TensorFlow v1/v2, Theano
Languages
- Italian: Native
- English: Advanced
Research
- NLP, Question Answering, RAG
- Seq2Seq, Transformers, LLMs