Large Language Models
Investigating the fundamental capabilities and limitations of large language models, with a focus on improving their reasoning abilities and reducing hallucinations.
Simone
Conia.
Founder @ Principled Intelligence
Half-Italian, half-Japanese. Investigating how to control frontier AI systems and make them more trustworthy.
41°54′ N · 12°30′ E
About me
I am the Founder & CEO of Principled Intelligence (opens in a new tab), where we work to make AI systems more trustworthy through evaluation and governance. My expertise spans Artificial Intelligence, Natural Language Processing, and Large Language Models. Previously, I was an Assistant Professor at Sapienza University of Rome and a Research Scientist at Apple. My work focuses on advancing the capabilities of language models and developing innovative NLP solutions.
Author of over 40 publications in top-tier conferences (including ACL, EMNLP, AAAI, IJCAI, NAACL), I have contributed significantly to the field of AI and NLP. My research has been recognized with multiple awards, including the Outstanding Paper Award at EMNLP 2024 and NAACL 2021. Earlier in my career, I was also a Distinguished PC at IJCAI, Honor Student at Sapienza, a CyberChallenge.IT podium winner, and winner of the Google Startup Workshop.
Awards &
recognition
-
Outstanding Paper
EMNLP 2024
-
Outstanding Paper
NAACL 2021
Distinguished PC
IJCAI
Honor Student
Sapienza University
3rd place
CyberChallenge.IT
1st place
Google Startup Workshop
Collaborations
Research collaborations and projects
Apple
Adobe
Amazon
Nvidia
Research interests
Multilingual AI
Developing robust AI systems that can effectively handle multiple languages, with particular attention to low-resource languages.
Retrieval Augmented Generation
Enhancing language models by integrating external knowledge retrieval mechanisms to improve factual accuracy and reduce knowledge gaps.
At Principled Intelligence
At Principled Intelligence, my research focuses on small language models (SLMs) for trustworthy AI: building specialized guardrails, evaluating their reliability, and making research usable through open model releases.
Our releases include ScopeGuard, a family of 4B-parameter models for multilingual scope and policy classification, and ClaimExtractor, which turns conversations into structured claims and intents for downstream fact-checking and auditing. I emphasize reproducible evaluations: clear experimental setups, comparisons on public benchmarks, and attention to the trade-offs between accuracy, latency, and deployment cost.
Explore our open models on Hugging Face
Selected publications
Towards Cross-Cultural Machine Translation with Retrieval-Augmented Generation from Multilingual Knowledge Graphs (opens in a new tab)
KG-MT introduces a novel end-to-end approach that integrates multilingual knowledge graphs into neural machine translation via dense retrieval, enabling significant improvements in translating culturally-nuanced entity names compared to state-of-the-art systems.
Increasing Coverage and Precision of Textual Information in Multilingual Knowledge Graphs (opens in a new tab)
M-NTA is a novel unsupervised approach that combines Machine Translation, Web Search, and Large Language Models to automatically generate high-quality multilingual textual information for knowledge graphs, significantly improving coverage and precision for non-English languages.
Unifying Cross-Lingual Semantic Role Labeling with Heterogeneous Linguistic Resources (opens in a new tab)
This paper introduces a unified model for cross-lingual Semantic Role Labeling that learns to map heterogeneous linguistic formalisms across languages without word alignment or translation, enabling robust and simultaneous annotation with multiple inventories.
Experience & education
Founder & CEO
Principled Intelligence (opens in a new tab)
Leading the development of tools to evaluate AI agents, identify failures, and govern their responses, helping organizations deploy AI more responsibly.
Assistant Professor
Sapienza University of Rome
Leading research in Natural Language Processing with focus on Large Language Models, Multilingual NLP, and Retrieval Augmented Generation. Teaching graduate courses in Computer Science.
Research External Collaborator
Apple
Collaborating on research projects related to natural language understanding and generation, focusing on improving the performance of language models in real-world applications, especially in multilingual contexts.
Research Scientist Intern
Apple
Conducted research on multilingual language models, focusing on enhancing their understanding and generation capabilities across various languages. Developed novel methodologies for improving knowledge-related question answering tasks across multiple languages.
Ph.D. in Computer Science
Sapienza University of Rome
Specialized in Natural Language Processing and Machine Learning. Dissertation on multilingual language understanding and generation.
Get in touch
I'm always interested in discussing research collaborations and innovative projects in NLP and AI.
Contact Information
Connect on LinkedIn (opens in a new tab)
Rome, Italy
Research Interests
- Large Language Models
- Multilingual NLP
- Retrieval Augmented Generation
- Evaluation
- Machine Learning