Nosible, the AI-driven alternative data provider, has launched World V2.0, an enhanced global event database designed to help institutional investors extract structured intelligence from the vast quantities of unstructured information published online. The release also underpins Atlas, the company’s recently introduced knowledge graph, which connects organisations, individuals, products and locations through historically traceable relationships.
The company crawls approximately 300,000 sources across 95 languages, ingesting between 10 million and 20 million web pages daily. World transforms this material into classified events, while Atlas maps the relationships identified within them, enabling investors to investigate how commercial, organisational and geopolitical connections evolve over time
From Alternative Data to Structured Events
Nosible’s approach has its roots in CEO Stuart Reid’s earlier career as a quantitative researcher, when he experimented with applying deep learning techniques to financial market data. Using structured datasets from providers including Bloomberg, FactSet and Capital IQ, he found that deep neural networks generally underperformed simpler models such as linear and logistic regressions. His subsequent experiments with alternative data included using Microsoft’s Bing Search API to uncover company information that was not yet appearing in mainstream financial data feeds.“That idea stuck with me: it’s not the model that counts, it’s the data, and whether or not you have different data that can lead you to a differentiated opinion,” he tells Market & Alt Data Insight. “That’s actually the genesis of any kind of alpha in financial markets.”
That experience eventually led Reid to develop Nosible’s own web search infrastructure, which now provides the foundation for World. The system identifies events across the broader information landscape, rather than limiting its scope to predefined financial instruments or companies.
“We don’t start with a singular view or dimension that we look at the data through, like finding all the stocks and then extracting the events related to those stocks,” he explains. “That’s not how we do it. We take all the data that’s published on a given day and use the features in the search engine to connect documents together. We look at the people they’re talking about, whether they have the same meaning and whether they’re discussing the same thing. Then the events emerge from the data, from people talking about the same thing at the same time. The key is corroboration. We need to see multiple documents from independent third parties talking about the same thing at the same point in time, and that’s what gets picked up and processed as an event.”
Once identified, events pass through an enrichment pipeline that extracts named entities and their relationships, classifies developments according to detailed taxonomies and assigns sentiment, materiality and forward-looking indicators. World V2.0 adds improved entity recognition, multilingual processing, more precise classifications and enhanced ticker mapping, making the resulting information more accessible for investment analysis.
Connecting Events Through Atlas
Atlas extends this approach by organising the extracted entities and relationships into a point-in-time knowledge graph. It currently encompasses approximately three million people, 2.4 million organisations, 700,000 products, 500,000 locations and 100,000 geopolitical entities, with around 35 million recorded facts.
These include supplier and customer relationships, employment connections and geographical associations. Each relationship is linked to supporting events and underlying source documents, allowing users to examine the evidence and establish when particular information became available.
The historical dimension creates opportunities to analyse changes in corporate networks, including developments that may not be immediately apparent from company-specific news or financial disclosures.
Says Reid: “You can go back in time and say, what did the supplier graph look like for Nvidia in 2022 versus 2018 versus 2026? How has that changed? You can imagine strategies based on organisational complexity, where the number of relationships being added to a company at each point in time is indicative of the amount of business they’re doing. Maybe that’s a signal that a company is becoming increasingly relevant in its sector and is worth buying. These are the kinds of strategies that we’re enabling firms to build.”
Nosible’s historical records extend back approximately 30 years, although Reid estimates that the most usable data covers the past 15–20 years because corroborating older events becomes increasingly difficult. Preserving the timing of information is particularly important for quantitative research, where historical backtesting must distinguish what investors could have known at the time from information established subsequently.
Establishing Trust in Global Information
Processing information at this scale introduces challenges around accuracy, provenance and source reliability. Nosible evaluates source credibility in relation to the subject being covered, rather than relying exclusively on fixed rankings of publishers.
“How reputable a source is depends on the topic they’re talking about,” notes Reid. “When the BBC is talking about current affairs, that’s an authoritative signal. When the BBC is talking about cancer treatment, maybe it’s less authoritative than a specialist publication. We do have mechanisms for scoring trustworthiness, but we think about source trustworthiness differently. We don’t have a singular ranking where the BBC is number one and the Financial Times is number two. It’s much more dynamic. We also actively filter out content that we flag as disinformation or content farms, and we provide customers with metadata about the sources, including who’s behind them, how long they’ve been publishing and what they usually write about.”
The company also deliberately includes sources such as Chinese state media, recognising that differences in how events are reported across countries and languages can themselves provide useful signals. Nevertheless, corroboration across multiple publishers does not necessarily establish factual accuracy, particularly where reporting originates from a common source.
Institutional Applications and AI Integration
Nosible primarily targets systematic and multi-strategy investment firms with the infrastructure to incorporate large-scale datasets into their research and trading environments. According to Reid, customers have developed strategies using the data across global equities, currencies and global macro, although specific performance results have not been disclosed.
The company is also exploring applications in prediction markets, including discussions with firms interested in using its intelligence to identify potential new markets.
World, Atlas and the underlying search infrastructure are accessible through APIs, with bulk data delivery available via Amazon S3. Nosible is also preparing an MCP interface to allow AI agents to interact with its datasets through natural-language-driven workflows.
As investment research becomes increasingly automated, the ability to connect current developments with historical relationships could expand the range of information available to systematic strategies and AI-assisted analysis. The question for Nosible, and for the investment firms adopting such technology, is how effectively that broader intelligence can be translated into reliable, differentiated signals.
Subscribe to our newsletter
