Alexander F. Spies

I work on making language models safer and more reliable. During my PhD I published four papers on causal world models & interpretability and received multiple AI-safety grants. At Epic Games I design pipelines that fine-tune and evalute frontier LLMs on low-resource data. I'm now looking to bring that blend of research rigour and production know-how to teams shipping reliable, scalable AI.

Thesis: Interpretable Representations in Artificial Neural Networks

• Research on object-centric representations and interpretability of LLMs
• Advisors – Prof. Alessandra Russo & Prof. Michael Shanahan

• Thesis: Learning World Models in the Animal-AI Environment
• Independent project: Neurosymbolic Learning & Neurally Weighted dSILP

• Completed graduate-level courses as an undergrad, alongside research

• Thesis: AI for the Automated Diagnosis of Atrial Fibrillation

Research‑engineer intern focused on large‑scale fine‑tuning of LLMs on low-resource languages.

• Implemented finetuning pipeline for local as well as cloud-based training (UnSloth, SageMaker, etc.).
• Deployed evaluation suite with W&B sweeps, vLLM serving and multiple LLM APIs.

Led independent research groups on language model behaviour and interpretability.

• Developed research agenda on mechanistic interpretability for understanding maze-solving LLMs.
• Trained transformers models and Sparse Autoencoders and developed interpretability pipelines.
• Managed 9 researchers across 2 projects resulting in 2 workshop papers and a best-poster award.

Mechanistic analysis of Transformers trained on maze‑solving tasks.

Investigated non‑local thresholds in pixel detectors; co‑authored JINST publication.

Exclusion analysis of Higgs decay channels in MSSM.

• Provided technical guidance on AI Safety Research to 8+ Research Fellows

• NeurIPS, ICLR, ICML, AAAI, UAI, Artificial Intelligence Journal

• Led technical coursework for Deep Learning, ML Math, Data Structures & Algorithms, and Python
• Engineered GPU-backed autograding pipeline for 120+ students using Otter Grader and Paperspace

• Evaluated research proposals for alignment‑focused projects