Applied AI research for autonomous aerial systems, focusing on multi-object tracking, real-time perception pipelines, and deployment-oriented validation.

Key Responsibilities & Contributions

• Designed and optimized multi-object tracking algorithms for UAVs using deep learning and sensor fusion approaches.

• Built and integrated real-time AI pipelines on embedded/edge systems using Python, C++, and GStreamer for high-throughput aerial data processing.

• Trained and refined object detection and tracking models using PyTorch and TensorFlow for dynamic environments.

• Conducted empirical testing and performance profiling (latency/FPS/robustness) to validate reliability in real-world conditions.

• Collaborated across disciplines to translate research into deployable solutions, aligned with operational and ethical requirements.

• Ensured robustness and compliance with security protocols for public-sector / defence-grade deployment contexts.

• Contributed to technical documentation and research outputs within AIML.

Lead the Vietnamese AI team in R&D, deployment, and building an AI backend module from scratch to automatically detect danger zones, send warnings/alerts if no-safety helmet is detected, etc for construction management applications.

Key responsibilities

• Conducted R&D to integrate new object detection, instance segmentation and classification algorithms into AI platform.

• Built services (MQTT and RabbitMQ) to establish communication between the designed AI module and other systems (on-premise, web platform).

• Built a full MLOps lifecycle.

• Integrated and deployed AI module into different systems (edge, on-premises, and cloud) for production.

Achievements

• Successfully built an inference pipeline that can run/deploy many algorithms such as object detection, instance segmentation, and classification models on various platforms (edge devices, on-premise PCs, Kubernetes clusters) with encryption consideration. 

• Successfully built training orchestration service with the self-hosted ClearML server, Kubernetes cluster. Training tasks are scheduled based on task’s priority to run on limited number of Kubernetes resources. 

Led the CoreTech Algorithm Development team in advancing wearable health technologies through applied research and algorithmic innovation.

Key Responsibilities

• Directed research and development to enhance the precision and performance of core biometric algorithms, including heart rate monitoring, step counting, calorie estimation, and SpO₂ measurement.

• Led the integration of a blood pressure estimation algorithm into the next-generation smartwatch platform, aligning with product roadmap and technical feasibility.

Key Achievements

• Successfully optimized all major health features to meet Google Wear OS Health Service standards, ensuring robust performance and regulatory alignment.

• Delivered the most accurate experimental results for blood pressure estimation, significantly reducing error margins and contributing to product readiness for commercial deployment.

Researched and developed AI models for medical applications. Utilized the latest technologies to register medical images in different phases, which assist doctors in diagnosing faster and more conveniently.

Key responsibilities

• Developed point-cloud-based medical image registration to support registering liver CT/MRI images captured at different phases.

• Built an end-to-end AI model to support registering five different kinds of transformation in only one deep learning network.

Achievements

• Developed a point-cloud-based model that can register two liver image volumes within a second with GPU computation and maximum 6 seconds with CPU computation.

• Built an AI-based registration model that supports different registration types, including rigid, affine, similarity/scaling, translation, rotation. 

Researched and developed robotics algorithms on top of the latest technologies of top conferences such as CVPR, ECCV, ICML, etc.

Key responsibilities

• Conducted R&D on a novel Augmented Reality guiding robotics system (Solpen).

• Conducted R&D on an Augmented Reality System applied in medical applications.

• Improved existing deep learning-based bin picking and packing systems.

Achievements

• Registered the novel AR guiding robotics system for patent.

• Successfully secured investment from two partner companies to integrate my new invention into their products.

• Built a deep learning-based system that can grasp unknown objects for industrial robots.

Research and develop AI models for medical applications.

Key Responsibilities 

• Conducted R&D to develop deep learning models for heart vessel segmentation from MRI images.

• Conducted R&D to develop deep learning models for skin mole detection and classification for early detection and treatment.

• Conducted R&D to develop deep learning models for video-based heart rate estimation for rehabilitation and fitness exercises.

• Conducted R&D to develop deep learning models for action recognition from video analysis for recognizing nurse’s actions while taking care of bedsore patients at Hsiao Hospital (Taipei, Taiwan).

Achievements 

• Successfully built a skin mole detection backend AI module to assist patients in predicting the early potential of mole cancer.

• Successfully built and tested surveillance camera-based heart rate estimation at Hsiao Hospital.

• Built video action recognition models for classifying CPR actions.

• Successfully built a heart vessel segmentation model with a high dice score up to 97%. 

Enhance the quality of the X-ray image of the chest by detecting breath, using regular / depth cameras (Kinect, Softkinect, Asus Xtion). 

Designing a system to produce the Indirect Evaporative Cooler (research about CNC Machine, PLC Direct Logic 205 of KOYO, Mitsubishi robot RV-14UHC-SA12, Driver ServostarCD, and DDR servo motor of Kollmorgen).