Works
Selected works across AI research, software & systems, and security — with code where it's public.
AI & Research
-
Python 76% Jupyter Notebook 20% Shell 3% 1 star 0 forks AGPL-3.0 Updated
Ho Chi Minh City AI Challenge — Multimodal Video Retrieval 2024–2025
Multimodal data intelligence platform for interactive video search and analysis, built for the Ho Chi Minh City AI Challenge (2024 & 2025 qualification rounds). Combines visual, textual, and temporal signals to retrieve target moments from large news-video corpora.
-
Flower Classification with ResNet-50 & Contrastive Learning 2024
Image-classification study on the Flowers-102 dataset for the Fellowship.AI challenge — a ResNet-50 backbone combined with a contrastive-learning pipeline, with analysis of class distribution, training dynamics, and an inference demo.
-
Vietnamese Text Classification with PhoBERT
Fine-tuning Vietnamese pretrained language models (PhoBERT, with a Longformer variant for long documents) for domain text classification — covering tokenization strategy, long-input handling, and evaluation of the final model.
-
Hopfield Network Plays Wordle
An associative-memory approach to Wordle — encoding the answer vocabulary into a Hopfield network and letting energy minimization recall candidate words from partial letter constraints, wrapped in a playable game interface.
-
Jupyter Notebook 89% Python 11% 0 stars 0 forks Updated
Banking Intent Classification with Fine-tuned LLMs
Parameter-efficient fine-tuning of compact language models for banking customer-intent classification, using the Unsloth training stack for fast LoRA adaptation on a single GPU.
-
Beer Detection with YOLO
Real-time object detection for beer products — a YOLO detector trained on custom data and packaged behind a simple user interface for running detections on images and streams.
Software & Systems
-
Maze Game — Pathfinding & Q-learning
A maze game doubling as an algorithms playground — procedural maze generation, side-by-side visualization of classical pathfinding, and a Q-learning agent that learns to solve the maze through reinforcement.
-
Swift 45% C++ 24% TypeScript 18% 1 star 0 forks Updated
SwiftUI 2048 with C++ AI
The 2048 game for iOS written in Swift/SwiftUI, with a C++ engine that plays the board automatically — native UI on top of a performance-critical search core bridged into the app.
-
TypeScript 55% Rust 21% CSS 18% 34 stars 6 forks GPL-3.0 Updated
mx — Markdown Editor
A fast, lightweight Markdown editor for the AI era, built with Tauri 2 and Rust — native performance and a small footprint with a modern TypeScript front end.
-
Go 31% TypeScript 28% JavaScript 21% 0 stars 0 forks MIT Updated
USRing — Embedded Sensing System
End-to-end embedded system for the USRing hardware — firmware for the primary ESP32 board and its camera module, plus a Go web server that interfaces with and manages the devices.
-
Python 100% 1 star 1 fork Updated
HCMUS Course Registration Automation
Async Python client for the HCMUS course-registration flow — adaptive concurrency control and a circuit-breaker pattern to stay fast and well-behaved against an overloaded backend during registration windows.
-
Remote Control via Email
Controlling a remote machine through email — commands are parsed from a mailbox and dispatched to a client agent, with results returned by mail; a study in building an RPC channel over an unconventional transport.
Cybersecurity
-
TypeScript 50% Python 48% CSS 1% 0 stars 0 forks Updated
DreamHack Wargame Crawler
A robust crawler that discovers, downloads, and organizes DreamHack wargame challenges into a local practice library — resumable downloads and structured metadata for offline study.
-
Python 91% C++ 9% 1 star 0 forks Updated
CTF_Lib — Solver Utilities
Helper libraries and utilities accumulated from real CTF solving — reusable building blocks for setup and problem-solving across cryptography, web, and misc categories.
-
Python 74% Sage 9% Dockerfile 5% 0 stars 0 forks MIT Updated
CTF Archive — Crypto Challenge Hosting
Self-hosted archive of notable cryptography challenges from past CTFs — keeping great problems runnable and solvable after their competitions end.