I am an AI Engineer with 5 years of experience in Computer Vision and 2 years in NLP and Large Language Models, hold a Master’s degree in Applied Artificial Intelligence from Deakin University with a High Distinction average. I specialize in translating business problems into scalable AI systems, including LLM-powered applications and AI agents deployed in production environments. I work effectively both independently and in collaborative teams, and I’m known for a strong growth mindset—approaching challenges as opportunities to learn, improve, and continuously expand my skills while building impactful AI solutions.

• Developed a Generative AI pipeline combining LISA (Vision-Language Model) for automated segmentation and Stable Diffusion with DreamBooth for synthetic dataset generation, achieving 50% improvement in dataset quality for wildlife detection models.
• Developed and optimized YOLO-based models for wildlife detection, supporting conservation-focused AI applications in Australia.
• Reduced ecological dataset collection costs by leveraging generative data augmentation to overcome camera-trap variability.

Team Leader — Capstone Project

• Implemented a real-time sentiment analysis system processing customer communications using Python and AI models to extract themes and emotional tone.
• Fine-tuned LLMs on sentiment datasets, achieving a 20% improvement in F1-score for customer feedback analysis.
• Developed automated real-time dashboards to monitor customer sentiment and support rapid issue resolution.
• Tech Stack: Mistral 7B, LLM Fine-Tuning, Prompt Engineering, MongoDB, Streamlit

• Implemented and deployed YOLO-based ML models for traffic violation detection and electronic toll collection systems, processing 1M+ vehicles daily across Vietnam's national highway network.
• Optimized model inference using TensorRT and ONNX for edge deployment, achieving real-time performance of 30+ FPS on embedded devices.
• Delivered on-premises enterprise Agentic AI Sales Assistant solutions using Retrieval-Augmented Generation (RAG) with Llama-2-13B, vLLM, LangChain, and LangGraph.
• Built scalable vector database infrastructure using ElasticSearch for domain-specific knowledge retrieval.
• Architected a real-time Social Listening system crawling 10+ social platforms and processing 500K+ posts daily using Apache Kafka and Spark.

• Led team of 4 engineers to develop Deep Learning solution for automated defect detection in mobile display manufacturing, replacing manual inspection process.
• Deployed Transfer Learning-based UNet and YOLO models for marked region detection/segmentation on edge devices, achieving 95% detection accuracy.
• Reduced manufacturing costs by 80% through minimizing false positive rates and eliminating manual quality control labor.
• Optimized models for real-time edge deployment using TensorRT and embedded systems integration.

• Built an Agentic AI Learning Companion that transforms long lecture videos into structured, searchable knowledge with LLM-powered Q&A across course materials.
• Designed a production-ready AI architecture with FastAPI (API layer), LiteLLM (LLM gateway), NVIDIA NeMo Guardrails (safety), Langfuse (observability), and RAG using ChromaDB.
• Developed a data ingestion pipeline with Airflow to process lecture video/text data, generate embeddings, and store them in a vector database for retrieval.

Tech Stack: CrewAI, ChromaDB, LiteLLM, Redis, FastAPI, Airflow

• Integrated SAM2 with Ultralytics for automated frame annotation and trained YOLO11 models for player and ball detection.
• Applied ByteTrack for multi-object tracking to follow players across frames in real-time sports footage.
• Developed OCR models to recognize player jersey numbers for enhanced automated game analytics.
• Built post-analysis features including player-focused camera generation, heatmaps, and performance statistics for coaching insights.

Tech Stack: YOLO, ByteTrack, SAM2, Ultralytics

• Developed a comprehensive traffic monitoring system with vehicle detection, tracking, and attribute extraction (class, color, direction, license plate).
• Implemented DeepSORT for multi-object tracking and OCR for license plate recognition achieving 88% accuracy.
• Built search functionality enabling queries by vehicle attributes and automated video summarization reducing review time by 75%.

Tech Stack: YOLO, DeepSORT, OCR

• 📜 DeepLearning.AI Certificate: Agentic AI, AI Agents in LangGraph.
• 📜 Microsoft Azure AI Fundamentals (AI-900)
• 📜 Microsoft Azure AI Engineer Associate (AI-102)

• 🤖 Agentic AI Systems: Building multi-agent workflows and LLM-powered applications for real-world problems
• 🔍 Computer Vision Research: Generative AI for synthetic data augmentation and wildlife conservation applications
• ⚙️ MLOps & Edge AI: Optimizing and deploying models at scale with low-latency inference
• 📚 Continuous Learning: Staying current with the latest AI/ML research and techniques

I'm always interested in collaborating on data science and AI projects, especially those with social impact. Feel free to reach out!

⭐ Fun Fact: I speak English, Vietnamese, and Korean, and I love exploring how AI can bridge language barriers and create more inclusive technology solutions!