Skip to content
View NAMEisNOTvailable's full-sized avatar
1 follower · 0 following
  • Adelaide, South Australia

Achievements

Achievement: Pull Shark

Achievements

Achievement: Pull Shark

Block or report NAMEisNOTvailable

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
NAMEisNOTvailable/README.md

Hongyu Su

Cyber Security | LLM Security | AI Safety Evaluation

I am a Master of Cyber Security graduate from the University of Adelaide, currently based in Adelaide, Australia. My repositories focus on reproducible evaluation workflows, clear metrics, runnable security demos, and practical mitigation notes.

Featured Projects

Project Focus Evidence
LLM Safety Evaluation Multilingual LLM security evaluation Mandarin-English prompt-injection benchmark with refusal/compliance scoring and cross-lingual risk findings
LLM Defend Chinese prompt-injection dataset composer Coverage-balanced dataset generation with deterministic rules, deduplication, and Chinese-first adversarial prompt design
GuardedIM Security Assessment Secure messaging security assessment Reproducible protocol flaws, PoC demos, Python/Go test coverage, and mitigation notes for encrypted message handling
Medical MLP Classification Imbalanced medical classification TensorFlow workflow with automated checks, SMOTE, validation-based model selection, threshold tuning, and metric-aware reporting
RNN/LSTM Stock Forecasting Financial time-series evaluation RNN/LSTM forecasting with a persistence baseline, RMSE/sMAPE/MASE/RMSLE reporting, tested helpers, result artifacts, and automated result checks
CNN Image Classification Computer vision model comparison PyTorch workflow with automated checks, pretrained ResNet-18 fine-tuning, and an AlexNet-style baseline

Technical Strengths

  • Security: vulnerability assessment, secure communications review, threat modelling, web/network security, Linux, Burp Suite, Wireshark, nmap.
  • AI / LLM evaluation: prompt-injection testing, multilingual robustness analysis, benchmark design, taxonomy design, refusal/compliance scoring.
  • ML / data: Python, TensorFlow/Keras, PyTorch, scikit-learn, NumPy, model evaluation, experiment tracking, tabular and sequence modelling.
  • Web / operations: WordPress, WooCommerce, HTML/CSS, SEO content architecture, product catalogue structure, checkout and customer-flow improvements.

Additional Practical Work

I also built, maintain, and operate Brinelume, a live e-commerce website for bath, body, home gifting, custom orders, and wholesale enquiries. This work includes WordPress/WooCommerce setup, product catalogue and variation management, checkout and customer-flow improvements, policy pages, SEO-oriented copy, and ongoing site operations.

Pinned Loading

  1. llm-safety-evaluation llm-safety-evaluation Public

    Mandarin-English LLM safety evaluation project for prompt-injection testing, refusal/compliance scoring, and cross-lingual security analysis.

    Python

  2. llm-defend llm-defend Public

    Chinese-first prompt-injection dataset generator for LLM security evaluation, with deterministic generation and coverage-balanced taxonomy design.

    Python

  3. guardedim-security-assessment guardedim-security-assessment Public

    Secure programming case study for a Python/Go encrypted messaging prototype, covering RSA/AES-GCM workflows, vulnerability analysis, and mitigation notes.

    Python

  4. mlp-medical-classification mlp-medical-classification Public

    Medical classification experiments using MLP models, imbalanced-data handling, SMOTE, and metric-aware evaluation.

    Python

  5. rnn-lstm-stock-forecasting rnn-lstm-stock-forecasting Public

    RNN/LSTM stock forecasting experiments with a persistence baseline, tested helpers, result artifacts, and CI checks.

    Jupyter Notebook

  6. cnn-cifar10-classification cnn-cifar10-classification Public

    CIFAR-10 computer vision experiments comparing CNN architectures, augmentation strategy, training behaviour, and evaluation metrics.

    Jupyter Notebook