Biography

My work is animated by a structural question: when does the spectral and geometric character of a learned representation determine what a learning system can and cannot do — and what does that imply about when pipelines will fail?

My research pursues this from complementary angles. One preprint (arXiv, 2026) introduces the spectral saturation index S(K) = erank(Σ̂⁽ᴷ⁾_W)/K, a label-free stopping rule for few-shot label acquisition validated across 49 tasks and three backbones. The other (Research Square, 2026) provides the first controlled empirical documentation of verifier exploitation in NLI-guided iterative refinement.

Both investigations were conducted independently, alongside a full formal school curriculum, without access to institutional compute, an academic supervisor, or a research group. All code, methodology, and writing is my own. I build the systems I study from first principles, emphasizing rigorous empirical evaluation and theoretical grounding.