Summary
Alex Spies is a research engineer specializing in mechanistic interpretability and neurosymbolic AI, with eight years' experience bridging deep research and production systems. He completed a PhD at Imperial College London studying interpretable representations, sparse/object-centric structure, and transformer computations, and has demonstrated emergent world-model structure in maze‑solving transformers. At Epic Games he built finetuning, eval and serving infra (UnSloth, SageMaker, vLLM) and shipped robust LLM agents that handle 10k+ weekly queries for low‑resource language code generation. He combines exploratory reverse‑engineering of models with production-grade reproducible pipelines and lightweight infra for large‑scale experiments and GPU monitoring. Alex has a track record of open-source tooling and dataset maintenance from his UnSearch work and workshops, and brings a physicist’s analytic rigor from earlier HEP and sensor research to AI system design.
7 years of coding experience
Physics, Physics at University of California, Berkeley
MPhys Physics with Theoretical Physics, MPhys Physics with Theoretical Physics at The University of Manchester
Doctor of Philosophy Artificial Intelligence, Doctor of Philosophy Artificial Intelligence at Imperial College London
A2 Levels, A2 Levels at Backwell School
English, German