Roone Ruanaidh
Research on safe agentic workflows — and the engineering required to make machine intelligence verifiable and reproducible.
Investigating whether engineering around an LLM — closed vocabulary, deterministic tools, cross-model review, interpretability-informed training — can produce agentic action that is bounded, verifiable, and reproducible, and whether the same engineering stabilises agent performance across open and state-of-the-art models on the same work.
Sphere is the independent research program where this work lives: invariant infrastructure that makes it safe for any stateless intelligence to take autonomous stateful action.
timeline
papers
Drafts live in private repositories. This section will list peer-reviewed and preprint work as it becomes citeable.
essays
Long-form forthcoming — on agent verifiability, interpretability-informed engineering, and the architecture of bounded autonomy.
résumé
A full résumé will live here. For now: LinkedIn.