Experience

Work Experience

VinMotion, VinGroup

• Role: AI/Robotics Engineer (PermanentFull-time)
• Period: June 2025 – Present
• Location: Hanoi, Vietnam
• Stack: ROS 2, Python, PyTorch, LangGraph, Tool Calling, VLA, VLM, Gazebo
• Bullets:
  • Developed embodied task-planning modules using multi-agent workflows and Vision-Language Models for autonomous task execution, and decision-making in humanoid robots
  • Built and deployed model-based and model-free 6DoF object pose estimation modules to improve robotic grasping accuracy
  • Developed an instruction-following data generation and benchmarking pipeline for VLN training, automating dataset creation and evaluation workflows
• Projects:
  • Task Planning for Embodied AI → /projects/task-planning-embodied/
  • 6-DoF Object Pose Estimation → /projects/pose-estimation/
  • Instruction Following Datagen & Benchmark → /projects/instruction-following-datagen-benchmark/
• Key Skills: ROS 2, LangGraph, VLM, VLA, PyTorch, Tool Calling, Humanoid Robotics, Gazebo

AIS, VMO Holdings

• Role: AI Engineer (Contract full-time remote)
• Period: May 2025 – Sep 2025
• Location: Hanoi, Vietnam
• Stack: Python, FastAPI, ONNX, LangChain, LangGraph, LLM, NLP, ASR, TTS
• Bullets:
  • Built a Computer Vision Acceleration Pipeline cutting development-phase time cost by 50–70%
  • Warehouse Package Tracking with RF-Detr deployed in warehouse via surveillance cameras, automating storage processes
  • Food-Ordering Agentic Workflow with LangGraph deployed in kiosks to automate ordering process from customer interaction (STT/NLP) to order fulfillment and customer feedback collection (TTS/NLP)
• Projects: [hidden — company privacy policy]

Human Robotics Laboratory, NAIST

• Role: Research Assistant (Contract full-time)
• Period: Nov 2023 – Sep 2025
• Location: Ikoma, Japan
• Stack: Flutter, Python, Unity, Unreal Engine, ROS 2
• Bullets:
  • Developed an Android VR See-Through System for medical assessment robots supporting clinical observation
  • Designed an Unreal-Based VR Driving Simulator with integrated AI driving assistant for driver training and cognition evaluation (Hitachi Corp collaboration)
• Projects:
  • Personalized Voice-Based eHMI → /projects/personalized-voice-based-eHMI/
  • Android VR See-Through System → /projects/android-vr-see-through/
  • Unreal-Based VR Driving Simulator → /projects/unreal-vr-driving-simulator/
• Key Skills: Flutter, Unreal Engine, Unity, ROS 2, Generative AI, HRI, VR/AR

Viettel Big Data Analytics Center, Viettel Group

• Role: AI Engineer (Permanent Full-time)
• Period: Nov 2021 – Sep 2023
• Location: Hanoi, Vietnam
• Stack: Python, PyTorch, Triton, Docker, Kafka, ASR, VAD, NLP, LLM, Speaker Verification
• Bullets:
  • Built and deployed large-scale speaker verification for Viettel Call Center — serving millions of customers daily with >99% accuracy, significantly reducing manual workload (>50%), preventing fraud, and improving customer experience
  • Developed a voice quality assessment system for call centers to monitor and improve audio quality for speech datasets, leading to a reduction of manual review time by 80% and improving ASR performance through better data quality
  • Participated in VoxCeleb Speaker Recognition Challenge (VoxSRC) 2022, Vietnamese Language and Speech Processing (VLSP 2022)
  • Data preparation and implementation of LLM into the My Viettel chatbot platform
• Projects:
  • Speaker Verification System → /projects/speaker-verification/
  • Voice Quality Assessment → /projects/voice-quality-assessment/
  • LLM Chatbot — fine-tuned PaLM & LLaMA on company data (2022) [no page]
• Internal Contributions:
  • Tech Talk: Speaker Verification for Call Center — internal knowledge-sharing session
  • Mentor — Viettel Digital Talent Program 2022
• Achievements:
  • High-rank placement in VLSP (Vietnamese Language & Speech Processing) competition
  • Participated in VoxCeleb Speaker Recognition Challenge (VoxSRC) 2022 (Top 20)
• Key Skills: Speaker Verification, PyTorch, Triton, ASR, VAD, LLM Fine-tuning, Docker, Kafka

Viettel Digital Talents Program 2021, Viettel Group

• Role: AI Intern (Part-time internship)
• Period: Apr 2021 – Oct 2021
• Location: Hanoi, Vietnam
• Bullets:
  • Completed intensive AI & Data Science training under Viettel’s national talent initiative
  • Contributed to Vietnamese speaker verification system later deployed in production
• Projects:
  • Speaker Verification System → /projects/speaker-verification/
• Achievement: Selected as Top 10 candidate out of the entire Viettel Digital Talent Program 2021 cohort
• Key Skills: Python, Speaker Verification, Deep Learning, Data Science

Education

MEng — Human Robotics Interaction

• School: Nara Institute of Science and Technology (NAIST), Japan
• Period: Oct 2023 – Sep 2025
• Research: Personalized human-robot interface for APMVs using generative models tailored to passenger personality.
• Funded by MEXT Scholarship, Japan (2023).

BSc — Automation & Electrical Engineering

• School: Hanoi University of Science and Technology (HUST), Vietnam
• Period: Sept 2018 – Sept 2022
• GPA: 3.52 / 4.0 — Graduated with Distinction, 2nd place in Thesis defense committee.
• Thesis: Speaker verification system for telecommunication service.

Competitions

VoxCeleb Speaker Recognition Challenge (VoxSRC) 2022 — Top 20

• Task: Speaker Verification — Track 1 (closed, no external data) and Track 2 (open, with external dataset)
• Approach:
  • Model fusion: RawNet (raw audio input) + ECAPA-TDNN (spectrogram input)
  • Multi-loss optimization: AAM-Softmax (classification) + Proxy Anchor (metric learning)
  • n-fold data augmentation pipeline

Vietnamese Language and Speech Processing (VLSP) 2022 — Top 5

• Task: O-COCOSDA and VLSP 2022 - MSV Shared task: Multilingual Speaker Verification (Track 1: seen languages, Track 2: unseen languages, Track 3: cross-lingual)
• Approach:
  • Model fusion: RawNet (raw audio input) + ECAPA-TDNN (spectrogram input)
  • Pretrained self-supervised models: Wav2Vec 2.0 and WavLM as feature extractors
  • Multi-loss optimization for cross-lingual speaker representation

Data-Centric AI Competition, FPT 2021

• Task: Improve human detection with YOLOv5 using a data-driven approach only (no model architecture changes)
• Approach:
  • Data preprocessing: noisy label removal, annotation standardization, image quality enhancement
  • Ensemble with diverse augmentation strategies and hyperparameter sets
  • Knowledge distillation to reduce manual data evaluation overhead