AI/ML Engineer Computer Vision and DL Expert

TRUONG CONG QUOC THAI

AI ENGINEER

+84-941-***-*** **********.**@*****.*** https://github.com/quocthaj Ho Chi Minh City SUMARY

Final-year Artificial intelligence student and Computer Vision enthusiast, distinguished by a unique blend of hands-on Generative AI expertise and image recognition capabilities. I have a proven track record of building robust automated systems using Python and Deep Learning frameworks, successfully handling large-scale datasets, and optimizing model performance for resource-constrained environments. Building upon a strong foundation in Machine Learning and Deep Learning, I am currently expanding my research focus into LLM Agents and RAG architectures. My core objective is to integrate complex data processing capabilities with Agentic Workflows to build comprehensive, automated AI systems. My 3-year vision is to become a key AI Engineer within 'AI-first' organizations, architecting and deploying breakthrough AI solutions—from core models.

PROJECTS

AI Co-worker Engine — Gucci Group CHRO NPC 1/2026 - 4/2026 Team size: 1 Technologies: Python, FastAPI, Gemini 3 Flash, Groq LPU, LangChain, React.

• Role: Architected and implemented the core Dual-Engine LLM architecture, managing state tracking, semantic routing, and real-time prompt injection.

• Scope: Developed a real-time, persona-driven AI Co-worker system capable of adhering to strict business constraints, managing contextual memory, and interacting with external tools via function calling.

• Results:

+ Achieved ultra-low inference latency by implementing Model Cascading and a Semantic Router, bypassing heavy processes for low-intent queries while effectively triggering Google Embeddings RAG for high-intent policy retrieval.

+ Engineered an "Invisible Supervisor" layer to synchronously evaluate user intent, calculate rapport scores, and dynamically inject behavioral prompt directives (e.g., persona shifting) without breaking the conversation flow.

+ Optimized context window efficiency and system speed by completely eliminating heavy frameworks (like LangChain), developing a custom SlidingWindowMemory and internal state tracker from scratch using native Python. Vietnamese Invoice Information Extraction (OCR) 12/2025 - 3/2026 Team size: 2 Technologies: Python, PyTorch, VietOCR, OpenCV, MC-OCR Dataset, Label Studio.

• Role: Complete data engineering process, model fine-tuning, and backend API deployment for an automated invoice processing system.

• Scope: Architected an end-to-end OCR pipeline to extract structured information (Merchant, Total Amount, Date) from diverse, real-world Vietnamese retail invoices.

• Results:

+ Significantly improved character recognition accuracy for complex Vietnamese fonts and handwritten-like text by fine-tuning the Transformer- based VietOCR model.

+ Increased system robustness against low-quality, blurry, or skewed captures utilizing advanced OpenCV preprocessing techniques (deskewing, noise reduction, adaptive binarization).

+ Successfully deployed the entire extraction pipeline as a high-performance RESTful API using FastAPI and EasyOCR, enabling seamless integration and returning clean, structured JSON data for downstream applications. Vietnamese Text-to-Image Generation (Generative AI) 12/2025 - 4/2026 Team size: 2 Technologies: Python, Stable Diffusion (Latent Diffusion Models), PyTorch, Hugging Face, Google Colab/Kaggle.

• Role: Executed comprehensive data preprocessing for both images and Vietnamese captions, and architected the end-to-end model training pipeline.

• Scope: Fine-tuned a Stable Diffusion backbone via LoRA using a curated dataset of 8,000+ image-text pairs to bridge the linguistic and cultural gap for Vietnamese text prompts.

• Results:

+ Generated high-fidelity visual outputs with accurate semantic alignment to complex Vietnamese linguistic inputs and cultural nuances.

+ Reduced training time and prevented overfitting through rigorous hyperparameter tuning on high-compute cloud GPU environments

(Colab/Kaggle).

+ Built a modular, scalable GitHub repository enabling seamless integration and deployment within the Hugging Face ecosystem. TECHNICAL SKILLS

Progamming Languages Python, JavaScript.

AI Modeling & Architecture • Computer Vision: CNN, ResNet50, YOLO, CRNN (VietOCR), Image Classification & Segmentation.

• Generative AI: Stable Diffusion (Latent Diffusion Models), LoRA Fine-tuning, Prompt Engineering.

• LLMs & Agentic Systems: RAG (Retrieval-Augmented Generation), Agentic Workflows, Semantic Routing, LLM Function Calling, Prompt Engineering, LLM-as-a-Judge. Frameworks & Libraries • AI & Deep Learning: PyTorch, TensorFlow/Keras, Hugging Face, Groq SDK, Google GenAI.

• Backend & API: FastAPI, Flask.

• Data & Vision: OpenCV, EasyOCR, scikit-learn, NumPy, Pandas. Tools & Infrastructure • Databases: Vector Databases (Google Embeddings), MySQL, SQL Server.

• Environment & Tools: Docker, Git/GitHub, Postman, Label Studio, Roboflow, Jupyter/Colab EDUCATION

HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY 2022 - 2026 (Expected) Artificial intelligence

CERTIFICATIONS

Networking Basic (Cisco Networking Academy) 2025

Intro to Machine Learning (Kaggle) 2024

JavaScript Essentials 1 & 2 (Cisco Networking Academy) 2025

© topcv.vn

Contact this candidate