About
Nguyễn Đình Nguyên Bắc
AI Engineer · SDV & AI
I build production AI for software-defined vehicles — computer-vision and edge-AI pipelines, virtual ECU / vehicle simulation, and multi-agent systems. My background spans applied research in speech, vision, and NLP — including international internships at A*STAR (Singapore) and National Chung Cheng University (Taiwan) — and end-to-end engineering, from model training to shipping a working product.
I graduated in Data Science from the Industrial University of Ho Chi Minh City, and I now work as an AI Engineer at AutoNxt AI on software-defined vehicles — bringing real-time computer vision and AI onto automotive hardware and into virtual vehicle simulation. I care most about the engineering that turns a promising model into something dependable: the pipeline, the evaluation, and the scaffolding that makes a model safe to ship. I like problems that cross layers — training a model, wrapping it in an API, and getting it running on the device.
On the research side I've worked on code-switching speech recognition, video deepfake detection, and Transformer-based image captioning; on the product side, tool-using agents and a full multi-agent e-commerce system. Lessons from that work end up on the blog, and the projects themselves live on the projects page.
Experience
- Work on software-defined vehicles (SDV): production computer-vision / AI pipelines, edge AI on automotive hardware, and virtual ECU / vehicle simulation.
- Project Kaizenics — built KX-Solution, a domain-agnostic agentic AI platform (multi-agent orchestration), with an SDV application: virtual cars on RemotiveLabs ECUs with Android Automotive and CARLA.
- Built Baby Safety Reminder, a separate child-presence safety system, with digital.auto.
- Built a smart HR assistant (recruitment, payroll, employee services) using LangChain + the OpenAI API.
- Implemented tool-using agents with memory, retrieval, and a vector database over internal systems.
- Researched code-switching ASR for Southeast Asian languages.
- Built a hybrid Whisper-encoder + LLaMA-decoder model with Noisy Student Training; reduced WER 34% → 28%.
- Researched video deepfake detection with Transformer models.
- Authored a research report and presented at the International Student Research Symposium at CCU.
Education
GPA 3.48/4.0 (Good) · Coursework: Deep Learning, NLP, Computer Vision
Skills
Get in touch
Email me at nguyendinhnguyenbac@gmail.com, or find me on GitHub, LinkedIn, and Facebook.