Scholar
Xuan-Phi Nguyen
Google Scholar ID: HN8VxX4AAAAJ
Salesforce
Citations & Impact
Citations
816
H-index
14
i10-index
21
Publications
20
Co-authors
12
list available
Publications
Resume (English only)
Academic Achievements
- Published multiple papers at top-tier conferences including ICLR, ACL, and EMNLP
- Paper on unsupervised machine translation published at ICLR 2022
- Paper on linguistic structures and neural architectures published at ICLR 2020
- Led SeaLLMs project, which received significant attention from media and research communities
- SFR-RAG models praised by Salesforce CEO Marc Benioff and widely covered in media
Research Experience
- 2024–Present: Senior Research Scientist, Salesforce AI Research, USA – leading SFR-RAG (compact LLMs specialized for Retrieval-Augmented Generation) and SFR-DeepResearch (an autonomous deep research agent trained via reinforcement learning)
- 2023–2024: Senior Algorithm Engineer (Research), Damo Academy, Alibaba, Singapore – led SeaLLMs, the first open-source state-of-the-art LLMs for Southeast Asian languages
- 2022: Research Intern, Meta AI (FAIR), USA – researched speech-to-speech translation with synthetic data generation and augmentation
- 2021: Research Intern, Facebook AI Research (FAIR), USA (remote) – developed novel pseudo-parallel data mining techniques for unsupervised machine translation (published at ICLR 2022)
- 2019: NLP Research Intern, Salesforce AI Research, Singapore – studied linguistic structures' impact on neural model performance in NLP tasks (paper published at ICLR 2020)
- 2018–2019: Research Assistant, NLP Group, NTU, Singapore – researched document-level MT, discourse, phrase/tree-based methods, and unsupervised NMT (papers in ICLR, ACL, EMNLP)
- 2017–2018: Software Engineer Intern, Visa Inc., Singapore – built lightweight character-level CNN for text classification with 1000× less resource usage and 10× faster training
- 2016: Software Engineer Intern, Panasonic R&D Center Singapore – developed SVM-based signal classifier (94% accuracy) and assisted in Raspberry Pi robot project for real-time car system control (83.8% real-time accuracy)
