ml-kit
The classic algorithms, from scratch — for offline features. One file, zero dependencies, no model, no API key, no network.
Boundary. This kit is for ranking, similarity, clustering and simple trend. It is not a diagnostic or predictive engine, and nothing in it may be used to claim a clinical or medical capability.
What's in it
tfidf+search— ranked text searchNaiveBayes— a tiny classifierknn,kmeans,pca2— similarity, clustering, plottingid3+predict— a small rules classifierlinearRegression— a trend linemarkov+generate— statistical generation from a corpusgrammar— rule-based generation (rules, slots, weighted alternatives)
Usage and the boundary statement are in the README. Tests: node test-node.js.
Scan with any UPI app. Payee: Dr. Sohil Momin.