I'm a software engineering student at UNIMAS who likes working where security, machine learning and systems meet. I run Vixero Technology Enterprise, and most of what I do falls into a few areas:

• LLM post-training: supervised fine-tuning (SFT) and reinforcement learning (RL) for language models
• Network security research: detection methods that work on traffic behaviour instead of packet contents
• Insider-threat research: LLM-driven multi-agent simulation and pre-registered experiments (HBEE)
• Applied ML and software: from sign-language translation to game mods

AEGIS: Adversarial Entropy-Guided Immune System · arXiv:2604.02149 · 2026 cs.CR (primary), cs.LG (cross-list) · CC BY 4.0

A network intrusion detection approach that classifies traffic from flow-level timing and entropy statistics, without inspecting payloads. It uses a state-space sequence model with a Mamba-3 core. On the evaluation set in the paper it reports an F1 of 0.995 with sub-millisecond inference (about 262 µs); see the paper for the setup and limitations.

Dataset: AEGIS-Adversarial-Corpus (908K sequences)

HBEE: Human Behavioral Entropy Engine · arXiv:2605.07472 · 2026

A pre-registered multi-agent LLM simulation study of insider-threat detection. Its main result is a detection inversion: an adaptive, LLM-driven insider drew less peer suspicion than a randomly chosen innocent agent, the opposite of the pre-registered prediction. The paper also states explicitly where the findings do not generalize.

Languages     Python · Java · C++ · JavaScript
ML/DL         PyTorch · TensorFlow · MediaPipe · LLM post-training (SFT, RL)
Systems       eBPF · CUDA · Linux · nginx
Infra         Hostinger KVM2 · AlmaLinux · GitHub Actions
Research      arXiv · HuggingFace · ORCID · ResearchGate · Pre-registered experiments

• 🌐 vixdev.cloud
• 🔬 arXiv
• 🤗 huggingface.co/Vix0007
• 🎮 modrinth.com/user/Vix.

"For Humanity, From Humanity."

Kota Samarahan, Sarawak, Malaysia 🇲🇾