NEXT-GEN AI FOR CYBER- AND NEUROPSYCHOLOGY
Building systems for
I build agentic AI, ontology-backed simulations, and reproducible research tools that make complex human behavior measurable, auditable, and useful.
Three chronological lenses on the same trajectory. Move between my experience, manuscripts, and education without losing the frame.
FeedPilot, Ghent
FeedPilot is a browser-based AI agent that acts as a post-recommendation layer across major social media platforms. It analyzes the visible feed in real time and selectively engages with content to steer future recommendations toward the user's own goals and values.
Key skillsAgentic AI, browser extension architecture, cross-browser compatibility, real-time DOM and media analysis, privacy-first product design
Ghent Health Psychology Lab, Ghent University
At the Ghent Health Psychology Lab, I contribute to six research projects focused on developing personalized digital health applications and data-driven methods for cancer survivors and people living with chronic pain.
Key skillsPersonalized digital health, back-end development, LLM-assisted research, EMA, accelerometry, multilevel network modelling, systematic reviews, data engineering
Designed and implemented the back end of a personalized decision-support engine for physical activity among cancer survivors. It combines safety filtering with explainable recommendations based on health conditions, goals, values, and contextual factors.
Developed a reproducible EMA and accelerometry pipeline examining how pain, fatigue, stress, and motivation relate to physical activity. Multilevel network models reveal individual patterns obscured by group averages.
Developed an idiographic EMA network analysis of how pain, threat, attention, and goal engagement interact in daily life, combining person-specific models with multilevel benchmarks.
Developed an OSF-registered systematic-review pipeline for literature retrieval, screening, structured LLM-assisted coding, ontology-aligned embeddings, mixed-method synthesis, and reproducible manuscript generation.
Engineered a privacy-preserving pipeline that transforms chatbot and engagement data into auditable datasets, descriptive analyses, academic figures, and an interactive evidence explorer with data-quality controls and HAPA-based mapping.
Developing an LLM-assisted workflow for literature screening, evidence extraction, conceptual coding, and synthesis while retaining expert validation and source-level traceability.
University of Washington eScience Institute, Seattle
Participated in NeuroHackademy 2026, a two-week intensive program combining neuroimaging, data science, open science, technical training, and a collaborative hackathon, in which I contributed to four projects.
Key skillsAgentic AI, neuroinformatics, multimodal neuroimaging, explainable AI, reproducible pipelines, Docker, open science, scientific visualization
Developed and validated an ontology-driven multi-agent research engine that integrates multimodal neuroimaging features and clinical context to generate explainable phenotype predictions, with actor-critic validation, traceable evidence chains, and demonstrations across MRI, EEG, and stroke-lesion data.
Contributed to an open educational website for visual quality assurance in MRI, fMRI, and EEG preprocessing, combining annotated artifacts, practical inspection guidance, curated tools, and automated resource monitoring.
Developed a reproducible, Dockerized pipeline that turns functional connectivity matrices into approximately 3,900 interpretable features, with a Python CLI, automated validation, visual diagnostics, a feature dictionary, and an interactive notebook.
Designed, developed, deployed, and currently manage an interactive WebGL website exploring 50 years of neuroscience through eight forms of biological and artificial intelligence using PubMed data, semantic embeddings, 3D visualization, graph dynamics, and data-driven sonification.
Technology Enablers for Cognitive Warfare, Amsterdam
Selected for a two-day NATO research workshop on technology enablers for cognitive warfare, contributing ontology-based state-space modelling and LLM-assisted simulation perspectives.
Key skillsCognitive warfare, cyberpsychology, ontology engineering, LLM-based simulation, adversarial modelling, institutional resilience
IIS Biobizkaia, Bilbao
Designed COMPASS, an open-source multi-agent system for multimodal zero-shot phenotypic prediction across large neuroimaging datasets and HPC workflows. Final evaluation: 19/20.
Key skillsMultimodal data integration, LLM-based pipelines, explainable AI, HPC scheduling, Docker, academic research and communication
IIS Biobizkaia, Bilbao
Built a private full-stack agentic GraphRAG platform for biomedical knowledge interaction, retrieval-augmented generation, and ontology-based reasoning.
Key skillsGraphRAG, semantic search, ontology engineering, REST API design, full-stack development
Bosspaints, Waregem
Six years of operational work in production, logistics, infrastructure maintenance, process optimization, and cross-departmental coordination.
Key skillsLogistics coordination, technical maintenance, teamwork, operational problem-solving
Rotating scope of open-source manuscripts. Select a blip to inspect the work and its title page.
Specialization in Neuroscience · Ghent University
Thesis on adaptive idiographic modelling with ontology-based agentic AI (the PHOENIX engine).
Credit Contract · Ghent University
Master-level trajectory across neuroscience, engineering, data science, machine learning, and neurotechnology.
Credit Contract · Ghent University
Certificates in human-data research, FELASA Category C equivalent, artificial intelligence, and Python programming.
Theoretical & Experimental Psychology · Ghent University
Foundation in experimental design, statistics, and cognitive science.
What I actually build. Three project groupings, grounded in the repository methods and systems behind them.
Ontology-grounded LLM systems that use explicit schemas, agent roles, critic loops, and constrained simulation spaces.
Closed-loop ontology-based agentic AI for adaptive idiographic modelling, from complaint operationalization to HUA time-series analysis, target selection, and HAPA intervention mapping.
github.com/stvsever/ThesisMasterClinical Ontology-driven Multi-modal Predictive Agentic Support with orchestrator, executor, integrator, predictor, critic, and communicator agents for multimodal phenotype prediction.
github.com/stvsever/COMPASS-EngineAdaptive hierarchy-aware Shapley attribution for structured LLM classification, built to audit which feature domains and leaf features drive clinical or phenotypic decisions.
github.com/stvsever/aHFR_TokenSHAPMachine-readable ontology of 1,149 simulation design choices, 44 hard constraints, stratified configuration sampling, SHFS scoring, and XGBoost plus TreeSHAP protocol derivation.
github.com/stvsever/research_paper_on_protocol_for_human_simulation_designPrivate biomedical GraphRAG assistant for CompNeuro knowledge exploration and ontology editing, plus PECAN2.0 ontology-backed recommendation logic for physical activity in oncology contexts.
Behavioral measurement systems for ambulatory data, semantic construct validity, LLM rating validation, and experimental cognition.
Reproducible EMA and accelerometry analysis with 1,474 completed assessments, ENMO activity data, partial-pooling mlVAR, S-GIMME checks, and person-level heterogeneity estimates.
github.com/stvsever/research_paper_on_physical_activity_in_fibromyalgia_patientsTransformer-embedding workflows for clinical construct audits, including hedonic versus eudaimonic scale loading and ontology-aligned biopsychosocial semantic analysis.
github.com/stvsever/research_paper_on_eudaimonia_in_healthcare github.com/stvsever/SystematicReview_on_BioPsychoSocialModel_in_ChronicPainResearchValidation pipelines that compare LLM relevance estimates with user ratings, build correlation and confusion matrices, and expose where data quality or prompt design limits performance.
github.com/stvsever/MappingComparison_UsersLLMExperimental and proof-of-concept tools for measuring irrefutable reasoning, insight problem solving, and automatic detection of question-order bias in survey design.
github.com/stvsever/research_paper_on_unfalsifiability github.com/stvsever/GEDRAGSECO_PROOF_OF_CONCEPT_BOQSecurity and autonomy work across cognitive manipulation, feed control, cyber education, assessment practice, and AI-assisted learning.
Browser extension and cloud backend that score visible feed items against user-defined goals, explain each verdict, enforce safety gates, and keep platform engagement logic subordinate to user intent.
steeryourfeed.comSecurity-framed ontology-based multi-agent simulation of how profile traits, DISARM attack tactics, opinion targets, and exposure-network position shape susceptibility to political manipulation.
github.com/stvsever/research_paper_on_cybermanipulation_susceptibilityCognitive-security map of how emotional states amplify or attenuate cyber-relevant bias families, using affective component calibration, clustering, and mixed-effects modelling.
github.com/stvsever/research_paper_on_cognitive_biases_across_emotional_states_cyberrelevantSingle-page interactive course covering cryptography, networks, web security, cloud, ATT&CK, ATLAS, D3FEND, DISARM, incident response, threat intelligence, governance, exercises, and a mock exam.
github.com/stvsever/Cybersecurity_FundamentalsStatic adaptive cognitive-assessment practice games, plus a local AI language tutor with vocabulary generation, scenario comprehension, audio, recall tests, and study analytics.
github.com/stvsever/HireVue_Practice_Cognitive_Assessments github.com/stvsever/LearnLanguage03 · Open channel
I partner with researchers, founders, and policy teams who need AI that holds up under scrutiny, where human behavior, cognitive security, and clinical science meet.
