Name Intelligence
A name parsing engine powered by Foundry. Splits human names into structured components β deterministic, with calibrated confidence instead of a single guess.
Common names worldwide
The given and family names Foundry sees most often across its whole training corpus, each shown with its share of what Foundry has learned β not an official statistic. See a specific country for its own breakdown.
Given names
- Maria 0.59%
- David 0.44%
- Ana 0.41%
- Jose 0.36%
- Daniel 0.36%
- Carlos 0.33%
- Antonio 0.33%
- Omar 0.32%
- Sara 0.31%
- Andrea 0.3%
Family names
- Khan 0.72%
- Kumar 0.33%
- Garcia 0.31%
- Lopez 0.29%
- Rodriguez 0.29%
- Martinez 0.26%
- Hernandez 0.25%
- Silva 0.23%
- Gonzalez 0.23%
- Perez 0.23%
Explore by country
See the demo pre-set for any of 165 countries, each with its own well-documented naming conventions where they exist.
Traditional name parsing treats a 100% confidence
output as if the system understood the input, even when it didn't.
Foundry says so when it doesn't β a much lower confidence score and a
plain note that the input needs a closer look, instead of a false
100%.
How it works
-
Tokenize & classify
Splits raw input into tokens and classifies each one against known name-part roles: given name, family name, particle, title, and more.
-
Weigh by language & statistics
Lexicon and frequency data across languages resolve tokens that could plausibly be more than one role.
-
Score & explain
Every token β and the parse as a whole β gets an explicit confidence score instead of a single pass/fail verdict.
Common name parsing challenges
-
Cultural name order
Given-name-first isn't universal β some conventions put the family name first, and formatting alone doesn't say which.
-
Ambiguous name parts
The same word can be a given name in one language and a surname in another.
-
Multi-part surnames
Compound and particle-joined family names (von, de, van) don't split at the last space.
-
Spelling variations
The same name can be transliterated or spelled multiple valid ways.
-
Lost detail in plain ASCII
Stripping diacritics and non-Latin scripts erases the detail that distinguished the name in the first place.
-
Genuine typos
Some input isn't a different name at all β just a misspelling of a real one.
Handles
-
Titles & honorifics
Academic titles, honorific chains, and salutations.
-
Given names
Multiple given names and initials.
-
Particles & nobility
Separates particles (von, de, van) from nobility ranks.
-
International surnames
Compound and multi-part family names.
-
Calibrated confidence
A score that reflects real certainty, not a fixed 100%.
-
Structured warnings
Flags genuine ambiguity instead of guessing silently.
Use cases
- Simplify signup and checkout forms β split a single "full name" field into structured given and family names automatically.
- Get the salutation right β generate a correct "Dear Mr./Mrs. X" opening for outbound email instead of guessing.
- Catch typos early β flag a likely misspelling before it gets stored, with a suggested correction.
- Personalize with a calibrated gender signal, not a blind guess β see the dedicated demo.
- See which countries a given or family name is actually common in.
- Clean up contact data β normalize inconsistent name formatting across CRM or import sources.
Deployment options
-
π Simple API
A simple, hosted REST API is in development and not yet publicly available. Need programmatic access now? Get in touch.
-
π Self-hosted / on-premise
Runs entirely inside your own infrastructure β a compiled binary, zero external dependencies, no data ever leaves your servers, exactly how this demo itself runs. Available on request.
-
π Offline (embedded devices, etc.)
No internet connection required at runtime β no cloud inference, no external lookups, fully local and fully deterministic. Available on request.
Why Foundry
-
No randomness
Parse the same name once or a thousand times β the answer never changes.
-
No guessing
Every token gets its own calibrated score, not a single blind answer.
-
No third-party dependencies
No GPU, no Python, PHP, JS, or SQL at runtime β no external API call either.
-
Lightweight
Runs on Unix (macOS, Linux, FreeBSD) today, with no extra dependencies β the same portable codebase is expected to reach Windows, iOS, and Android too.
-
165 countries
Calibrated per-country name data, not one global guess applied everywhere.
-
Typo-aware
Suggests a likely correct spelling instead of silently accepting or rejecting a probable typo.
Bring Foundry to your own data
API access, on-premise deployment, an offline integration, or custom development for your data β get in touch with the details.
Early-stage β reach out and help shape what ships next.
Frequently asked questions
What does it return for a name?
A structured breakdown β given name(s), family name, titles, and particles β plus a confidence score for the parse as a whole and for each individual part.
Can it guarantee a name is parsed correctly?
No. Foundry reports a calibrated confidence score instead of a fixed 100%, and flags low-confidence results with a warning rather than guessing silently.
Does it detect gender?
Yes, based on given-name statistics, together with a confidence score β not every name has an unambiguous gender.
Is submitted data stored?
The submitted name and country are logged in aggregate (name, country, and a running count) for statistical analysis β never with an IP address, timestamp, or any other per-request identifier.
What happens with misspelled names?
If a likely correct spelling exists, Foundry surfaces it as a suggestion (see "Did you mean" in the demo above) without silently overriding the input.
Why not just split the name with a spreadsheet formula?
A formula like TEXTSPLIT only cuts on a delimiter β it
has no idea that "von", "Dr.", or "Jr." aren't part of the family
name, and breaks the moment the names in a list aren't all the same
format. Foundry classifies each word by its actual role instead,
whatever the mix.