We scrape filings, donations, contracts and the open web, then use LLMs to pull out entities and the links between them. The output is a map of influence you can actually query.
Microsimulation, input-output models and bespoke indices. We build the number when the official statistic does not exist yet, or arrives two years too late.
LLM pipelines that work through millions of documents (annual reports, registers, plans) and turn them into structured, searchable datasets. We have run this on supercomputers.
Public-facing databases, trackers and indexes. Care workers checking a sponsor, reporters chasing a lead, a union making a case. The dataset becomes a tool in someone's hands.
Millions of pages scraped to map the modern far right and trace its links into money and power, across America and Europe.

An LLM pipeline reading every UK annual report to surface confirmed risk events as they appear. Live, and updating.

Political donations linked to government contracts. Companies that gave to Labour took £138m back in deals. Launched in the Guardian.

An AI-augmented index of the Heritage Foundation's 900-page plan, so anyone can search what it actually proposes.

Labour's shift toward business donors from 2019 to 2024, drawn from twenty-one years of donation records.

An economic model of landlord returns across England, built for the Joseph Rowntree Foundation.

A searchable database of licensed care-visa sponsors, built with the Bureau of Investigative Journalism so migrant workers can find legitimate employers.

Arts-council funding by constituency since 2014, built for Equity. A union argument, made checkable down to the postcode.
Mapping the corporate connections of the UK's entrepreneurial far right, from Companies House outward.

The first thing we shipped. The UK workforce scored by Covid exposure, on national press within days and on ITV's Peston.

30 million job adverts tagged with LLMs on the Isambard supercomputer, built with the UK AI Security Institute.

A co-mention network built for the ITUC, the global trade-union body, surfacing the firms working against labour rights.

Runs the unit. Builds the pipelines and the platforms the rest run on. Has been here since JARI in 2020.
Method work and ML research, from exposure scoring to the harder modelling problems.
Builds the document-reading and entity-linking machinery behind GERM, RTBB and the network maps.
Twenty-one years of donation records, donor-contractor nexus, the political-data investigations.
The economics. Landlord returns, climate-driven food inflation, the bespoke indices.
Forecasting and nowcasting, including predicting regional poverty ahead of the official figures.
Most of our work starts with a conversation. If you have data that should be evidence, write to us.