Thanh Hai Nguyen

Thanh Hai Nguyen

AI Engineer & Researcher

Hands-on experience in developing and optimizing AI/ML solutions, with a strong foundation in deep learning, recommender systems, and LLM-based agents. Skilled in building ML pipelines, evaluating model performance, and improving AI system efficiency.
Nguyễn Thanh Hải

Experience

AI Engineer (Viettel Digital Talent) | Viettel Networks ↗

  • Skill-based LLM agents for telecom RCA, evaluating skill quality, adherence, and utility with LLM-as-Judge.
  • Continual skill evolution framework leveraging historical execution traces and automated validation.

Research Assistant | DSKT Laboratory, VNU-UET ↗

  • Research on recommender systems: counterfactual explanations, reproducibility, and benchmarking beyond Top-1.
  • Token-efficient multi-agent recommendation architectures with role-aware model allocation.

Education

VNU University of Engineering and Technology | Bachelor of Science in Information Technology

  • GPA: 3.75
  • Scholarship: EVN Scholarship (2025–2026); Vung Tuong Lai Scholarship (2026); Academic Encouragement Scholarship (Fall 2025)
  • Honor: UET Academic Excellence Award (2023–2026); VNU Student with Five Good Merits 2025

Publications

SIGIR 2026

From Top-1 to Top-K: A Reproducibility Study and Benchmarking of Counterfactual Explanations for Recommender Systems

Quang-Huy Nguyen*, Thanh-Hai Nguyen*, Khac-Manh Thai*, et al. The 49th ACM SIGIR Conference on Research and Development in Information Retrieval.

SOICT 2026 · Oral

Fewer Tokens, Smaller Agents: Role-Aware Allocation for Efficient Multi-Agent Recommendation

Hoang San Le*, Thanh Hai Nguyen*, Thi Cam Van Nguyen, Hoang Quynh Le, Duc Trong Le. International Symposium on Information and Communication Technology (SOICT 2026).

Projects

Multi-Agent ArchitectureToken EfficiencyPyTorchAgentScope

A token-efficient multi-agent recommendation framework replacing conversational coordination with a structured Plan–Work–Solve workflow, achieving over 10% gains in NDCG@3/5 while reducing input tokens by 62.9%.

LLM-as-JudgeSkill EvaluationContinual EvolutionViettel Networks

Investigated skill-based LLM agents for root cause analysis of telecom network incidents, building multi-criteria evaluation metrics and a continual evolution loop to refine operational skills.