Global Data Sources — Why More Data Alone Is Not Enough

CI's Data Infrastructure

Contextual Intelligence aggregates data daily from global public sources. These fall into five categories:

Altogether, the data corpus (ORSD — Open Regulatory Signal Dataset) comprises over 5,000 companies, ~50,000 regulatory signals, ~20,000 patents, and ~10,000 clinical trials. The data is orchestrated via Apache Airflow and updated daily.

The Illusion of Data Volume

More data sounds like better results. In practice: data volume without context is noise.

A simple example: the FDA database contains every 510(k) clearance since 1976 — over 200,000 entries. A keyword search for "PCR" finds ~5,000 of them. Which of these 5,000 are relevant for a manufacturer of real-time PCR kits in 2026? Which are obsolete (technologically outdated)? Which signal current market activity?

Without semantic context and temporal classification, the answer remains: "You'll have to figure that out yourself."

From Data Silo to Knowledge Graph

The key difference with CI is not the number of data sources — but how they are linked in a semantic knowledge graph.

Data from all global public sources is not simply collected in a database. Instead, it is modeled as a graph (Dgraph + PostgreSQL) in which every entity is connected to others:

• An FDA clearance is linked to the manufacturer
• The manufacturer is linked to its patents
• The patents are linked to the underlying technologies
• The technologies are linked to grant programs that address them

This structure enables graph traversal: instead of a flat list of results, the user receives a contextual path — for example, "FDA clearance for manufacturer X → its patent Y → funding focus Z, which matches the technology".

Open Data by Default

All raw CI data is openly licensed as ORSD (Open Regulatory Signal Dataset) (ODbL v1.0 for the database, CC0 1.0 for content). Anyone can use, download, and build upon the data for free.

The strategy: open raw data establishes ORSD as the standard in the regulatory intelligence space. The more users that consume the data, the more improvements flow back into the dataset through ODbL reciprocity.

CI's premium product lies not in the data itself — but in the contextual assessment, semantic matching, and prioritization. Additionally, CI runs the entire pipeline on self-hosted foundation models (NVIDIA DGX Spark) — no data egress, no US cloud dependency, fully GDPR- and EU AI Act compliant.

Conclusion

Global public sources are a good starting point. But only semantic linking in a knowledge graph turns this data into usable business intelligence. Those who merely collect data have plenty of information — but little insight.