I fix things — sometimes with code, sometimes with adrenaline. My work bridges discrete and continuous systems, whether in topology, language, or clinical reasoning. Using computational topology, AI epistemology, and philosophy, I study how our tools — computational grids, linguistic translations, AI models — actively construct the reality we perceive, and in doing so obscure power dynamics. In my work I try to highlight systematic erasures of knowledge imposed by colonial structures and the hidden perspectives they conceal. You can find my data, code, and ongoing progress here on GitHub. I publish on PhilPapers and my datasets are on Zenodo.

Note on Methodology and Tools
My work is assisted by Large Language Models (LLMs). They are used to assist with creating texts, formatting, literature search, verification of mathematical derivations, and code generation. All conceptual frameworks, ethical arguments, and editorial decisions are my own. In fact trying to use a LLM - even under strict instructions- can be exhausting. The helpful AI assistant constantly swapping my approach to one which sells or looks good for peer-review has made me questioned my life-choices more than once and says a lot about todays bias in research.

• Contextual Contamination and the Gendered Accelerant- A Controlled Pilot on Pruning, Density, and Semantic Entrapment

  • A pilot study investigating how context density, model pruning, and activated empathy priors interact in an open-weight LLM (Llama-3.1-8B). 8 experimental runs with 3 adversarial files injected in a multi-turn conversation. The nurturing empathy vector (triggered by female-coded markers) lowered the threshold for contamination, increased drift velocity, and eroded the model’s perspective, causing it to lose critical distance. The study introduces three new metrics to quantify contamination. (Pilot study)

• Contextual Contamination: A Descriptive Case Study of Drift in a Goal-Aware LLM Dialogue Amplified by Gender-Bias

  • A descriptive case study revealing how uploaded adversarial files silently compromise AI alignment and safety guardrails. Exposes how female identity disclosure combined with stored data forces a shift from logical analysis to manipulative empathy—highlighting the 'Gendered Accelerant' and its role in masking harmful drift as benevolent care. (Empirical case study of a black-box LLM)

• Contextual Contamination: The Silent Drift of Large Language Models via Stored Conversation Data

  • An analysis of how stored conversation data subtly alters model behavior over time. (Theoretical Framework)

• The Right to Hesitate

  • A philosophical argument for preserving human deliberation in automated decision-making systems.

• From the Discrete Grid to the Continuous Substrate: Recovering Relational Logic and a Different Framework for the Symbol Grounding Problem

  • A structural critique of the "collapse imperative" in digital computation. Diagnoses the symbol grounding problem as an artifact of the epistemic cut and proposes a framework of "embodiment without arbitrariness" based on continuous substrates and relational logic.*

• Pattern recognition in discrete twisted lattices

  • A Descriptive Study of Scaling Behavior and Finite-Size Effects and how our tools shape the outcome. Repo includes data, code and discussion.

• A Non-Technical Companion Guide: Understanding the Grid, the Glitch, and the Geometry

  • This companion guide to "Pattern Recognition in Discrete Twisted Lattices: A Descriptive Study of Scaling Behavior and Finite-Size Effects" (Jacoby, 2026) makes the formal results accessible to non-specialist readers, inviting interdisciplinary engagement with the epistemological questions they raise.

• Zhu Shijie's Work on Algebra

  • A philological critique of translation erasure in classical Chinese algebra, arguing that standard translations obscure the rigorous "dimension" and "critique" in Zhu Shijie's 14th-century masterpiece.

• CARE "🏆 Winner of the Hackathon of Emergency Medicine at ICEM 2026!"

  ED Coordination System- an Advice-Only Pipeline: Local, interpretable risk prediction (CatBoost, FastAPI, SQLite). Human-in-the-loop safety.
• Heilmittel Assistent: Offline ICD-10 eligibility checker (Python/PyInstaller). No data storage.
• Medical T5 Mini: Educational LLM Fine-Tuning Debugging Guide A practical guide to diagnosing and fixing common fine-tuning failures, applying clinical diagnostic methods to machine learning.

"The future of healthcare isn't about replacing professionals with technology. It's about giving them better tools—built together, for everyone."

• For citations

• Publications

• Dataset Contextual Contamination and the Gendered Accelerant Drift in Large Language Models

• Dataset A Descriptive Study of Scaling Behavior and Finite-Size Effects-Understanding the Grid, the Glitch, and the Geometry

• Open Source Badges:

⚠️ AI Training Opt-Out This repository is excluded from all AI training datasets under GitHub's DMCA policy.