Open to opportunities

Duc Hoang Tien

@duchoangtien

Message

AI Engineer and Researcher building efficient on-device generative AI, RAG agents, and multimodal vision systems.

Vietnam

Message

What I'm looking for

I’m excited to work on efficient, on-device generative intelligence—building RAG and agentic AI systems that are optimized for edge deployment. I want a team focused on real-world productionization and measurable latency/memory wins.

I’m an AI Engineer and Researcher with 8+ years of hands-on experience in generative AI, large language models, and multimodal vision-language systems. I build efficient, production-ready solutions that meet real latency and memory constraints—especially on edge and mobile-class hardware.

I’ve led research-to-deployment transitions, designing and deploying RAG pipelines and agentic AI systems with tool use, semantic routing, and reflection-based self-correction. In production, I develop LLM/VLM inference systems and optimize models through quantization (TensorRT, ONNX, INT8/FP16) and knowledge distillation, including llama.cpp-based on-device inference.

Most recently, I led a team of 7 as an AI Scientist & Team Leader, delivering an agentic RAG architecture for natural language querying across a 1,000+ camera surveillance network and mentoring engineers with structured workflows, evaluation frameworks, and MLOps monitoring. I’m now looking to contribute to Qualcomm AI Research’s mission of advancing efficient, on-device generative intelligence.

Experience

Work history, roles, and key accomplishments

Current

AI Scientist & Team Leader

Current

FPT Software

Sep 2024 - Present (1 year 9 months)

Led a team of 7 engineers across computer vision and generative AI, deploying an agentic RAG system enabling natural-language querying across a 1,000+ camera surveillance network. Optimized multi-modal video understanding for resource-constrained edge hardware using quantization and inference tuning, and established MLOps monitoring and evaluation workflows.

Agentic RAG Computer Vision Semantic Routing MLOps CI CD

CEO & Founder (On-Device AI)

Edge Vision

Mar 2023 - Sep 2024 (1 year 6 months)

Founded an offline physical AI startup delivering on-device generative AI and computer vision without cloud dependency, productizing internal LLM/RAG and vision-language pipelines for SME use cases. Integrated multi-modal AI with IoT/AIoT platforms and owned model selection, quantization strategy, and inference optimization to meet device constraints.

Physical AI On Device Generative AI Offline LLM RAG Model Quantization AI Integration Computer Vision

AI Solution Architect

Amizen Labs

Sep 2021 - Sep 2024 (3 years)

Developed and benchmarked multi-modal LLM/VLM systems for real-time threat analysis, deploying quantized inference with TensorRT-LLM and llama.cpp. Built a production RAG pipeline with semantic routing, MonoBERT reranking, and reflection-based self-correction (Milvus/FAISS/LanceDB), and served models via Triton on AWS/GCP with NVIDIA TAO fine-tuning; generated synthetic data using NVIDIA Omniverse

LLamacpp TensorRT Llama.Cpp RAG With Semantic Routing MonoBERT Reranking Triton Inference Server Stable Diffusion

Computer Vision Engineer

Chkincam

Sep 2021 - Dec 2021 (3 months)

Implemented DeepStream-based pose estimation and real-time human activity classification using TensorRT. Built low-latency P2P video streaming using aiortc/HTTP for edge deployment.

Pose Estimation TensorRT AIORTC HTTP

AI Team Leader - Video Analytics

Wesmart

Mar 2020 - Sep 2021 (1 year 6 months)

Led development of production-grade video analytics models for face recognition, multi-object tracking, and vehicle management with a focus on real-time inference efficiency. Deployed optimized models to Jetson Nano/Xavier NX using TensorRT and DeepStream, reducing latency via FP16/INT8 quantization, and set up MLOps monitoring and CI/CD with Prometheus/Grafana.

Multi Object Tracking TensorRT FP16 INT8 Quantization MLOps CI CD

Computer Vision Engineer

FPT AI

Jun 2018 - Mar 2020 (1 year 9 months)

Built OCR data augmentation and synthetic data pipelines supporting model training. Trained and fine-tuned detection and recognition models in PyTorch and TensorFlow and deployed them via Triton Inference Server and TensorFlow Serving.

OCR Data Augmentation Synthetic Data Generation PyTorch TensorFlow Triton Inference Server

Education

Degrees, certifications, and relevant coursework

Vietnam National University – University of Engineering and Technology

Bachelor of Science, Electronics and Communication Engineering

2014 - 2018

Grade: Distinction (GPA 3.26/4.0)

Activities and societies: Vietnam Women’s Union Scientific Research Scholarship (2017); Students in Scientific Research Award (2018); Students in Scientific Research program presentation (2018).

B.Sc. in Electronics and Communication Engineering at Vietnam National University’s University of Engineering and Technology (2014–2018). Graduated with distinction (GPA 3.26/4.0).