Namor is a pronounceable-name generator that uses corpus-based phonotactics rather than random letters. Instead of producing gibberish like “xqztl,” it learns which sound combinations are plausible by analyzing real name patterns.

How it works:

  1. Syllabification — breaks training names into onset, nucleus, and coda
  2. Frequency extraction — counts how often each pattern appears in real names
  3. Weighted sampling — generates new names by probabilistically combining patterns

Trained on 43 language corpora plus 26 mythological name sets. Ships as a Vite SPA with a standalone HTML build (~11KB, zero dependencies) that pre-computes frequency tables at build time. All computation happens in the browser — no server, no API calls.

TypeScript, Vite, corpus data from MatthiasWinkelmann/firstname-database. MIT licensed.

Written with Claude / oh-my-pi running open models.