Leo Nguyen
AI Engineer · Trivita AI
I'm Leo (Nguyễn Huy), an AI engineer at Trivita AI building medical AI in Vietnamese. I own the full data lifecycle — from EMR ingestion through synthetic generation to versioned datasets — and the post-training and evaluation loop on an on-prem multi-node H100 cluster, with partner access to GB200 NVL72. My work sits at the intersection of Vietnamese NLP, real-world healthcare data, and large-scale compute, under international mentor lineage from Taiwan AI research.
What I'm working on
MedVita
Vietnamese medical history-taking assistant: ICD-10 VN normalisation, SOCRATES-structured history-taking per department, tiered safety guardrails, integrated with a central hospital's HIS.
LLM evaluation platform
started the company's self-hosted eval platform (FastAPI, Celery, Redis, Langfuse, DeepEval, DeepTeam): medical G-Eval metrics, LLM-as-judge calibration, red-teaming.
Post-training loop
eval → targeted data generation → SFT/DPO with LoRA/QLoRA on an on-prem multi-node H100 cluster, day-scale iteration cycles.
Experience
- AI EngineerTrivita AISep 2025 – present
- AI ResearcherAIMAJun 2024 – 2025
- Multi-Agent VQA Research Team LeadVLAI2023 – 2024
- Time-Series Research Team LeadAI VIET NAM2022 – 2023
Education
- B.Sc. Information TechnologyVAMK, Finland
- Professional AI TrainingAIO 2024
Stack
- Python
- FastAPI
- vLLM
- TGI
- TensorRT-LLM
- DVC
- MinIO
- Label Studio
- Langfuse
- W&B
- Docker
- Kubernetes
- GitHub Actions