About Irish Origins

Irish Origins is an AI-powered genealogy platform built to help the global Irish diaspora explore, search and connect historical Irish records. We bring together civil registration data stretching back to 1864 and census household records from 1901, 1911 and the newly released 1926 Census of the Irish Free State — integrating them through a suite of data engineering and machine learning techniques that go well beyond simple keyword search.

The platform was designed from the ground up to treat genealogy as a data problem. Behind the search interface lies a substantial data pipeline: millions of records ingested, cleaned, standardised and cross-referenced across six distinct datasets, with the goal of surfacing connections that would take a researcher hours to find manually.


Record datasets

Birth records1864–1924

Civil registration records of births across all Irish counties, including child name, district, mother and father surnames.

Death records1901–1974

Civil registration records of deaths including cause of death, age at death, district and county.

Marriage records1845–1949

Civil registration records of marriages including groom and bride forenames and surnames, district and date.

Census 19011901

Full household census returns for every county in Ireland, with occupation, religion, language, birthplace and family relationship data.

Census 19111911

Full household census returns including additional fields on marriage duration, children born and children living.

Census 19261926

The first census of the Irish Free State, covering 26 counties. Released to the public in April 2026 — one hundred years after it was taken.


How records are linked

Connecting the same person or household across different record types and census years is the core data science challenge at the heart of Irish Origins. Historical Irish records are inconsistent in spelling, formatting and completeness — a person recorded as Murchadha in 1901 may appear as Murphy in 1911 and as Murphey on their birth certificate. Bridging these variations requires more than exact matching.

Surname clustering

Every surname in the platform is mapped to a cluster of related variants drawn from a curated Irish surname index covering over 9,700 spellings across 2,000 clusters. A single cluster groups together the English spelling, the Irish-language form, prefix variants (O', Mc, Mac, Ní, Nic) and known phonetic variants — so a search for Murphy will also search Murfey, Morphey, Ó Murchadha and all related forms simultaneously. Users can also choose individual variants manually.

Phonetic matching with Soundex

For cross-census household linking, exact name matching alone is insufficient. Irish Origin uses Soundex phonetic encoding to group surnames that sound alike even when spelled differently. This is applied after exact matching as a second-tier approach — yielding an additional 128,000 1901-to-1911 household links and 77,000 1911-to-1926 links that exact matching alone would have missed.

Irish name normalisation

A custom normalisation function strips Irish prefixes (O, Mac, Mc, Fitz, Ní) and all non-alphabetic characters before comparison, allowing O'Brien, OBrien, O Brien and Ó Briain to match the same underlying root. Diacritics (fadas) are handled using the PostgreSQL unaccent extension, so Ó and O are treated as equivalent across all search and linking operations.

Census-to-census household linking

Households are linked across the 1901, 1911 and 1926 censuses by matching the head of household name against the same DED (District Electoral Division) and county in the adjacent census year. Over 493,000 1901-to-1911 household links and 296,000 1911-to-1926 links have been established — covering approximately 40% of all households in each census. Linked households are surfaced directly on the census record page, enabling a researcher to follow a family across three census snapshots spanning 25 years.

Probabilistic civil record linking

Birth and death records are linked probabilistically using a combination of name matching and age consistency. For a birth record, potential death records are identified by matching surname and forename, then filtering by the constraint that the death year minus the age at death should fall within three years of the birth year. The same logic runs in reverse from a death record to surface potential birth registrations. Results are split by county — same-county matches appearing first as higher-confidence candidates.

AI-generated household narratives

Each census household record is accompanied by an AI-generated description, produced on demand by Claude (Anthropic's large language model). The model is given the full household data — names, ages, occupations, religion, language, birthplace and family relationships — and prompted to write a historically informed 3–4 sentence description grounded in the social context of Ireland in that census year. The 1926 descriptions are calibrated separately, reflecting the different data available and the unique historical significance of that census as the first of the Irish Free State.


How to use the search

Enter a forename and/or surname in the search box. Filter by county and year range if needed. Results show record counts grouped by type — click any type to drill down into individual records. Click a result to open the full record page, including a household graph for census records, original document images where available, and potential linked records across other datasets.

Tick Search all name variations below the surname field to see all known spelling variants for that surname and select which ones to include in your search. This is particularly useful for Irish surnames with many historical spellings.