Founder of Informatic Edge, LLC
Full-stack data scientist specializing in bioinformatics and computational biology
Bioinformatics pipelines, database architecture, and cloud platforms built to serve real research data at scale.
Custom methods, model development, and the AI ops work that keeps them running after the notebook is closed.
Cosine metrics reproduce to four decimals on a second machine. Mapper metrics don't, and can't be expected to. The first post's anchor ρ turns out to be a single draw resting on about eleven documents; with a 25-seed estimator the two machines agree, and one of the first post's headline contrasts doesn't survive.
Cosine leaderboards rank by training objective, not corpus fit. Adding a Mapper-TDA stability layer and linguistic anchors changes which embedder wins. A fourteen-encoder roster spanning both training-regime axes (contrastive vs MLM, general vs biomedical). Results indicate the topology ranking is structured by training-regime + pooling-convention compatibility, not cosine benchmark performance.
The PI-as-entrepreneur analogy flatters both sides but obscures where the real risks, incentives, and accountability actually live.
New posts on bioinformatics, data science, and the odd self-hosting tangent.