Co‑Founder — TextPast
Building tools that turn historical archives into searchable, verifiable knowledge for cultural‑heritage institutions and private collections. textpast.com
Data engineer turning archives into knowledge a machine can answer for.
When a model can fabricate any portrait, any document, any voice, resemblance stops being evidence. What still separates a record from an invention is provenance — and provenance is an engineering problem. Every clip on this page was generated from a historical source; the discipline is in being able to trace it back.
We get lost in hallucinations… see the Portfolio & Writings to find a trail back to true sources of provenance.
Simon co‑founded TextPast to build RootReal — archival AI whose first promise is provenance.
RootReal answers questions about the past by reasoning over inventories, indices, and transcriptions — enriched through ontologies and linked to relevant external sources. A custom context‑augmentation pipeline grounds every answer in verified archival data, so each statement can be traced back to the record it came from.
It is informed by two decades of hands‑on archival practice — from handwritten‑text recognition and georeferencing to Linked Open Data — across roles in data engineering, digitization, and research.
See how it works in the RootReal presentation, Providing Provenance →
Building tools that turn historical archives into searchable, verifiable knowledge for cultural‑heritage institutions and private collections. textpast.com
Developed a natural‑language application that generates Linked Open Data from the Archives' inventories, indices, transcriptions, and external sources. A bespoke Rust pipeline feeds verified source data to a large language model, which answers user questions by pointing directly at the underlying records.
Initiated research into a machine‑learning Named‑Entity application for the digitized Indies / Indonesian newspaper collections on Delpher (National Library of the Netherlands) and Opac (National Library of the Republic of Indonesia).
Researched Margono Djojohadikusumo (1894–1978) — founder of the Indonesian National Bank and a leading indigenous expert at the colonial Office of Co‑operatives. The biography draws on long‑term analysis of the Javanese aristocracy's socio‑economic position via handwritten‑text recognition of VOC, colonial, and Republican archives.
Led the digitization of over 21,000 pages of Javanese manuscripts from private collections in Central and East Java. Handwritten‑text recognition and probabilistic indexing produced the first‑ever text‑recognition model of Javanese script. eap.bl.uk
Led the digitization, data‑mining, and cataloguing of the Kartini Collection — the most extensive private map and print collection in Indonesia.
Applied photogrammetry and photography to digitize and catalogue objects in the museum of the Hadiningrat court of Yogyakarta. kratonjogja.id
Led the digitization of over 30,000 pages of colonial notary records from Javanese towns — long thought lost, traced by the team in private collections. Data‑mining revealed unique insights into the early‑twentieth‑century colonial real‑estate market. colonialarchitecture.eu
Analysed and georeferenced thousands of Dutch colonial maps and prints from the Rijksmuseum, the National Archives in The Hague, and affiliated institutions. atlasofmutualheritage.nl
Built the database of diplomatic letters found in the daily records of the Dutch East Indies archives.
A thesis on the architecture of RootReal — how archival sources are modelled, enriched, and made answerable with provenance intact. One subchapter applies the architecture in practice: multimodal data analysis and mapping — including campaign mapping — across centuries of sources, from fifteenth‑century Samarkand to late colonial surveys, using Handwritten Text Recognition, Named‑Entity processing, and GIS to project them onto georectified early‑modern maps. The work is being continued at a more technical university; an update is forthcoming.
Graduated Cum Laude. Data Analytics · Analytical Skills
Methodology & Statistics, with a track in history. Graduated Cum Laude.
Selected work in archival data engineering — systems that turn four centuries of inventories, transcriptions, and maps into knowledge a machine can answer for, without ever losing the trail back to the source.
A bespoke Rust pipeline that grounds every language‑model answer in verified archival data — reasoning over inventories, indices, and transcriptions, enriched through ontologies and linked sources. Each statement it produces can be traced directly back to the record it came from. Provenance is not a feature bolted on at the end; it is the architecture.
A natural-language assistant that automates the creation of archival aids and answers visitors' questions about the collection — reasoning over inventories, indices, and transcriptions as Linked Open Data, with every answer pointing back to the record it came from.
Led the digitization of 21,000+ pages of Javanese manuscripts from private collections in Central and East Java. Handwritten‑text recognition and probabilistic indexing produced the first‑ever HTR model of Javanese script.
Led the digitization of 30,000+ pages of colonial notary records from Javanese towns — long thought lost, traced by the team in private collections. Data‑mining revealed unique insight into the early‑twentieth‑century colonial real‑estate market.
Analysed and georeferenced thousands of Dutch colonial maps and prints from the Rijksmuseum, the National Archives in The Hague, and affiliated institutions — projecting historical cartography onto modern coordinates.
How archival sources are modelled, enriched, and made answerable by a language model without losing the trail back to the record.
A doctoral study of multimodal data analysis and mapping across centuries of sources — HTR, Named‑Entity processing, and GIS projected onto georectified early‑modern maps.
Notes on the gallery above: generating motion from a documented original, and why the provenance record — not the result — is the deliverable.