Locai Labs launches UK synthetic persona dataset with NVIDIA

Locai Labs and NVIDIA unveiled Nemotron-Personas-UK, an open-source dataset of 1 million privacy-preserving synthetic personas meant to help developers build and test UK AI systems with less bias. The dataset is designed to reflect the UK’s demographic, geographic and socio-economic mix and will join NVIDIA’s broader Nemotron Personas collection next week during GTC Berlin.

Why it matters: - Nemotron-Personas-UK gives developers a large synthetic dataset to test AI systems against UK-specific population patterns without using personal or private data. - The dataset is intended to help reduce bias in models by reflecting differences across regions, cultures, languages, age groups, education levels and rural-urban populations. - The release matters for sovereign AI efforts in the UK because it is built to support local model development and evaluation.

What happened: - Locai Labs Limited announced Nemotron-Personas-UK in collaboration with NVIDIA. - The open-source dataset contains 1 million privacy-preserving synthetic personas grounded in UK demographic, geographic and socio-economic distributions. - The dataset goes live next week during GTC Berlin and will be added to the NVIDIA Nemotron Personas collection.

The details: - Nemotron-Personas-UK includes 1M personas, 6M persona fields and 18M contextual fields. - The dataset contains 0.59B tokens, including 0.23B persona tokens. - The data covers 372 local authorities across England, Wales, Scotland and Northern Ireland. - Population share in the dataset follows adult population proportions: England and Wales 889,311 records, or 88.9%; Scotland 83,163, or 8.3%; Northern Ireland 27,526, or 2.8%. - The dataset includes 8,999 unique first names and 23,807 unique surnames. - It was developed on a hybrid set of 18 NVIDIA GPUs using cloud and local compute, including Locai Labs’ Locai 1 AI computer. - The dataset is released under a Creative Commons Attribution 4.0 licence. - Locai Labs said the dataset aligns to real demographic and occupational distributions of UK populations, but no real person is represented. - The company said it also accounted for differences across the UK’s four nations, including census systems, educational qualifications and language distribution. - Locai Labs said Celtic language proficiency was factored into persona generation for use in language-capable AI models.

Between the lines: - Synthetic data is becoming a core tool for training modern large language models because it can be generated at scale. - Synthetic datasets can still skew toward a narrow set of profiles if they lack diversity, which can carry bias into AI systems. - Representative personas are a way to preserve diversity in generated data and better mirror real users. - The dataset extends NVIDIA’s Nemotron Personas Collection, which already includes France, Japan, Brazil and the USA, with Germany also set to join during GTC Berlin. - James Drayson, CEO of Locai Labs, said the company was chosen by NVIDIA to develop the UK dataset and hopes it will support a wider range of UK sovereign AI systems.

What's next: - The UK dataset will be available in NVIDIA’s Nemotron Personas collection next week. - Developers can use the dataset to train personalization engines, fine-tune models on culturally grounded data, test domain-specific systems in healthcare, law and finance, and simulate representative user interactions for conversational AI. - Locai Labs will continue positioning Locai 1 and its sovereign AI work around on-premises and edge deployment for enterprises.

The bottom line: - Locai Labs and NVIDIA are betting that privacy-preserving synthetic personas can help UK AI systems become more accurate, less biased and more locally relevant.