cuong.nm
Open to freelance· Web · LLM · Mechanism modeling (math / FEM)

Hi, I'm

Cuong (Henry) Nguyen

AI Engineer · Vibration-Based Fault Diagnosis & Edge AI

Mechatronics engineer from HUST. Most of my time goes into AI for industrial fault diagnosis and backend for LLM products. I build websites for personal and small-scale projects (you can see some products below).

Cuong (Henry) Nguyen

Skills

Programming Languages

Python, C/C++, Matlab & Simulink

Web Development

HTML, CSS, JavaScript, TypeScript, React, Next.js, Tailwind, GSAP, next-intl

Software & Tools

Autocad, Solidworks, VS Code, Office

Research Fields

Signal Processing for Fault Detection, Vibration Analysis, PHM, RUL Prediction, Explainable AI, Transfer Learning

Education

  1. Bachelor of Engineering, Mechatronics

    Hanoi University of Science and Technology • Hanoi, Vietnam • 2021 - 2025

    Good Degree · English B2

Experience

  1. Backend Developer

    Aladin • in 2026

  2. PHM Researcher

    ITD Lab — Hanoi University of Science and Technology • 2025 - Present

Research

Fault Diagnosis of Planetary Roller Screws via Digital Twin + Transfer Learning

Master's Researcher · 24 months (2026–2028)

Problem

Planetary Roller Screws (PRS) are high-precision actuators used in robotics, aerospace and CNC machines, but real-world failure data is extremely scarce, preventing direct training of conventional deep-learning models.

Approach

  • Analytical model. Hertzian 3-body contact + 6-DOF Lagrangian dynamics; compute the characteristic frequencies RPF / RSF / NBF.

  • Digital Twin. Roller FEA in ANSYS → export MNF → rigid-flex integration in MSC Adams. Sweep fault parameters (pitting / crack / wear / preload loss), validated to <10% error vs. analytical frequencies.

  • Simulation dataset. 48 scenarios × 100 s @ 20 kHz, 1 s sliding window with 90% overlap; physical features (kurtosis RPF/NBF, VMD) + 1D-CNN deep features.

  • Transfer Learning. Compare three strategy families: instance-based (TrAdaBoost+KLIEP), feature-based (MMD/CORAL/DANN-lite), adversarial (CDAN/MCD) to reduce the simulation↔real domain gap.

Work

ITD Lab Website

Bilingual lab site, every animation handwritten

2026

Designed and built itdhust.com for the Intelligent Technical Diagnostics Lab at HUST. Six bilingual pages (EN / VI): a publications archive driven by BibTeX, an auto-translated news feed, member directory, admissions, and an events gallery. Every animation is hand-written in GSAP — scroll reveals, a cursor spotlight, 3D-tilt cards, magnetic buttons, an aurora background, and a partner marquee.

My Role

Built it solo: design, frontend, content schema, SEO, deployment.

Highlights

  • Publications page reads a single BibTeX string. Paste an entry, it parses and groups by year, no manual list to maintain.
  • News and event captions are written in Vietnamese; English is generated on the fly via Google Translate and cached forever.
  • Vietnamese names are reordered to Western form automatically (PGS.TS Nguyễn Trọng Du → A/Prof Trong-Du Nguyen), with academic titles mapped to English.
  • Sitemap with hreflang, robots.txt, ownership verified on Google Search Console.

Stack

  • Next.js 16
  • TypeScript
  • Tailwind v4
  • GSAP
  • next-intl
  • Vercel
Learn more

Aladata

Conversational Text-to-SQL (Vietnamese)

A Vietnamese-language interface for asking business-data questions without writing SQL. The system parses the question, generates the query, runs it on ClickHouse, and returns the result. Three modes: single questions, follow-ups, and switching context between topics.

My Role

Backend Engineer. I own the Memory subsystem (Redis + Postgres + Graphiti/Neo4j) and the FastAPI service that wires the pipeline together.

Highlights

  • Built and own the full Memory stack: Redis for short-term context, Postgres for long-term storage, Graphiti/Neo4j for the knowledge graph.

Stack

  • Python
  • FastAPI
  • LangGraph
  • Redis
  • PostgreSQL
  • Neo4j
  • Qdrant
  • ClickHouse
  • Docker
  • LangFuse

Personal Projects

  • Real-Time Fault Diagnosis on Edge

    Edge AIVibration

    Softmax Regression and Random Forest deployed on edge hardware for standalone real-time fault diagnosis from raw vibration signals.

    View on GitHub

  • Random Forest & Logistic Regression

    Signal Processing

    Vibration-based fault diagnosis pipeline with time–frequency analysis, advanced signal preprocessing and optimized classical models.

    View on GitHub

  • CNN Transfer Learning

    CNNTransfer Learning

    CNN-based transfer learning to adapt fault diagnosis models across operating conditions with limited labeled target-domain data.

    View on GitHub

  • Transformer-Based Fault Diagnosis

    Transformer

    Conv-Transformer and Vision Transformer for time-series and time-frequency data, achieving accurate diagnosis of complex mechanical faults.

    View on GitHub

  • Softmax Regression from Scratch

    PythonC++

    Softmax-based neural network implemented from scratch (no ML/DL libraries) and deployed in C++ for low-level, embedded AI execution.

    View on GitHub