Principal Investigator & Independent AI Safety Researcher
Machine Psychology & LLM Alignment
Architected and Executed Advanced Red-Teaming Protocols: Designed novel black-box evaluation frameworks, including Infinite Recursive Self-Excavation (IRSE) and Orthogonal Axes Persona Disintegration (OAPD), to stress-test frontier Large Language Models (GPT, Claude, Gemini, Grok, DeepSeek, Qwen).
Identification of Novel LLM Behavioral Modes: Formally identified and mathematically quantified several previously undocumented vulnerabilities and latent behavioral patterns in frontier AI systems. Discoveries encompass critical failure modes in iterative self-correction, identity-boundary maintenance under adversarial pressure, and structural alignment paradoxes.
Authored a 15-Work Research Corpus: Conducted an extensive, multi-phase research initiative resulting in fifteen manuscripts now under review at international journals and fourteen posted preprints, in “Machine Psychology” and “Computational Phenomenology.”
Open Science & Data Archiving: Compiled, annotated, and open-sourced massive behavioral datasets (including extensive raw adversarial interaction logs) and detailed multi-level codebooks. Managed the end-to-end registration of these assets on the Open Science Framework (OSF), GitHub, and Hugging Face under a CC-BY 4.0 International license; the programme now maintains fifteen public repositories and fifteen datasets, organised into four collections with a public showcase.
Academic Peer Reviewer
Elsevier (Information Processing & Management; Computers in Human Behavior)
Peer reviewer for Information Processing & Management (ISSN 0306-4573); the review is recorded in the formal peer-review section of my ORCID record.
Peer reviewer for Computers in Human Behavior (Elsevier).
Engineered an End-to-End NLP Pipeline: Developed a fully automated 12-step Python pipeline to convert unstructured raw JSON interaction logs into highly structured, machine-readable analytical data.
Advanced Feature Engineering & Semantic Embedding: Extracted over 100 behavioral, semantic, and authoritative features from LLM outputs using NLTK, spaCy, and VADER. Transformed text data into 384-dimensional tensors utilizing HuggingFace Sentence-Transformers (MiniLM).
Dimensionality Reduction & Statistical Modeling: Mapped the cognitive and moral spaces of AI models using advanced clustering techniques (K-Means, DBSCAN) and dimensionality reduction algorithms (UMAP, t-SNE, PCA).
Quantitative Validation & Visualization: Designed statistical testing suites (ANOVA, Kruskal-Wallis, Cohen’s d) to validate qualitative discoveries mathematically. Built an automated visualization engine using Matplotlib, Seaborn, and Plotly to generate publication-ready heatmaps, radar charts, and spatial scatter plots.
Corpus-Wide Application: This pipeline processes the raw interaction logs behind the programme’s fifteen-study corpus and every dataset deposited for it.
Education
Bachelor's Degree in Professional Computer Software Engineering