A zero-runtime-dependency TypeScript parser for choosing a greeting-safe first name from a website Full Name field, designed for customer and lead follow-up messages.

import { getFirstName } from "first-name-parser";

getFirstName("Austin");                   // "Austin"
getFirstName("Austin P. Smith");          // "Austin"
getFirstName("Smith, Austin P.");         // "Austin"
getFirstName("Dr. Austin Smith Jr.");     // "Austin"
getFirstName("Austin Smith, Jr.");        // "Austin"
getFirstName("Austin Smith, MPH");        // "Austin"
getFirstName("Dr.Austin Smith");          // "Austin"
getFirstName("Mary-Jane Smith");          // "Mary-Jane"
getFirstName("J. Austin Smith");           // undefined
getFirstName("Dr. Smith");                 // undefined

Until an npm package is published, install directly from GitHub:

npm install github:Serbyte-Development/first-name-parser

General-purpose person-name parsers try to identify every component of names from many cultures and input formats. This package solves a narrower messaging problem.

First Name Parser is intentionally optimized for a common website workflow: someone submits a Full Name field and the business later wants to send a natural follow-up such as Hey Austin, thanks for reaching out....

The goal is to return a name only when it is safe to put directly into that greeting. If the parser cannot confidently determine a greeting name, getFirstName() returns undefined and the caller chooses its own fallback, such as omitting the name entirely.

Its core assumptions are explicit:

• First [Middle...] Last is the default order for an unpunctuated US form value.
• Last, First [Middle...] is recognized as listing order.
• First Last, Jr. remains given-name-first; a suffix after a comma does not automatically flip the name order.
• A clear single submitted name is returned as the greeting name. A one-letter submission is treated as an initial. Ordinary two-letter alphabetic values such as AJ, MJ, ML, or Li remain usable because initials and short nicknames can be natural greeting names; capitalization alone does not lower confidence.
• Leading initials are preserved by parseFirstName() but withheld by getFirstName(). J. Austin Smith must not silently become Austin, because that can turn a middle name into the greeting name.
• Common titles, suffixes, credentials, post-nominal role acronyms, odd whitespace, apostrophes, and hyphenated names are handled without a general production name database.
• Obvious non-name values such as email addresses, URLs, placeholders, numeric/user-like submissions, and organization-shaped values are withheld.
• Multi-person submissions such as John & Jane Smith, John Smith or Jane Smith, or slash-separated names return the first listed given name. A clear selected first name remains high confidence; a selected initial such as J. remains medium.
• The parser never falls back to a likely surname just to produce a value. Dr. Smith returns undefined instead of producing an awkward greeting such as Hey Smith.

Returns a high-confidence greeting name as string | undefined.

getFirstName("Jordan P. Smith");
// "Jordan"

getFirstName("J. Smith");
// undefined

const firstName = getFirstName(submission.fullName);
const message = firstName
  ? `Hey ${firstName}, thanks for reaching out...`
  : "Thanks for reaching out...";

Returns the parser candidate plus structural metadata. Unlike getFirstName(), this lower-level API exposes medium- and low-confidence candidates so callers can intentionally apply a different policy.

parseFirstName("Smith, Jordan P.");
// {
//   firstName: "Jordan",
//   confidence: "high",
//   format: "family-first"
// }

confidence is high, medium, or low. format is empty, single, given-first, or family-first.

For ordinary customer messaging, prefer getFirstName(). It only returns high confidence results. parseFirstName() is for callers that want to inspect or override that threshold.

Confidence describes how strongly the parser believes the candidate is safe as a greeting name. Multi-person submissions deliberately choose the first listed person, so another person in the field does not lower confidence by itself. medium commonly represents an initial-only candidate. low means the candidate is structurally ambiguous or the input does not contain a trustworthy greeting name.

The parser was developed against two complementary fixture sets:

• 804,225 generated US name forms using 53,615 first names and 156,621 last names from the 2020 US Census name datasets. The restored benchmark fingerprint-matches the historical 99.9669% exact / 99.9918% weighted result on the historical parser. The current parser reaches 99.9745% candidate accuracy on the same reconstructed cases; getFirstName() is intentionally stricter because it can abstain.
• 2,352 labeled person-name records extracted from the probablepeople project test/training corpus. parseFirstName().firstName reaches 98.64% exact-match accuracy on that noisier corpus.

These public corpora are useful regression checks for candidate extraction and name structure. They should not be read as getFirstName() greeting-safe accuracy: that helper intentionally returns undefined for medium/low-confidence candidates that labeled name corpora may still count as answers. The product contract is to choose a safe greeting name for real website follow-up and abstain when that name is not trustworthy. See BENCHMARKS.md for methodology and limitations.

Human names are not universally inferable from a single string. Without punctuation, inputs such as Taylor Jordan can be structurally ambiguous. This parser deliberately uses the expected US website-form order instead of guessing from first-name or surname frequency.

This library is not trying to recover a person's complete legal given name. For greeting purposes, Wen Kai Li -> Wen can be a successful result even if Wen Kai is the person's full given name. If you must know the actual legal or complete given name, collect it explicitly.

It is not intended for international name-order detection, authoritative legal-name decomposition, identity verification, or extracting every name component. If your application needs full title/first/middle/last/suffix parsing, use a broader person-name library.

npm install
npm test
npm run typecheck
npm run build
npm run benchmark:public

The library has no runtime dependencies.

A private real-form benchmark can also be run with npm run benchmark:private when the ignored benchmarks/private/ corpus and manually reviewed ground truth are present locally. Private client data and labels must never be committed or published.

Developed & maintained by Serbyte Development.