A UK government-backed initiative set out to build a secure, national marketplace connecting cultural and heritage institutions with AI developers who need training data. The mission is significant: unlock value from decades of archives, collections and digitised assets across major national institutions, while ensuring rights holders retain control and see fair commercial return. The pilot phase runs on a commercial SaaS exchange platform, but the marketplace faces a familiar cold-start problem — the storefront exists, yet supply and demand are not visible to each other, and matching them by hand does not scale.
What Datatronika built
Datatronika designed and built a tactical intelligence layer that sits alongside the marketplace and attacks the cold-start problem directly. The system continuously scans dataset platforms, research publications and market signals to surface what AI developers are actively looking for, cross-references that demand against a growing catalogue of institutional collections, and proposes concrete matches — flagging which archives could serve which buyers, with clear rationale and confidence levels on every suggestion.
Architecture
The architecture is cloud-native and governed from day one. A document database with native vector search enables semantic matching across demand signals, buyer profiles and supply collections. A serverless ingestion layer pulls from external APIs on a schedule, storing raw and processed data through a structured bronze–silver–gold pipeline. Every match is written as a reviewable proposal, not an automated commercial decision — human judgement stays in the loop for anything rights-related or commercially material.
Constraint: sovereign-grade trust
A defining constraint shaped every design choice: the system will eventually hold confidential institutional and commercial data. Identity, access control and auditability were treated as first-class requirements from the earliest prototype — not retrofitted later. Managed identities replace credentials wherever possible, secrets live in a vault, and every component is scoped to the minimum access it needs. Architecture decisions were made with the compliance bar a sovereign, government-adjacent platform will eventually have to clear.
Outcome
The immediate outcome is a working prototype that turns a manual, spreadsheet-driven matching process into an automated pipeline the internal team can act on directly — surfacing dozens of live demand signals, structured profiles of prospective AI buyers, and a catalogue of institutional supply, all matched and ready for commercial review within weeks rather than months. It has already reframed how the project’s leadership thinks about the platform’s next phase.
Looking further out, the same intelligence layer becomes the foundation for a bespoke, sovereign version of the marketplace itself — a multi-year national platform, not only a document database and a set of functions. The prototype exists precisely to prove that architecture and de-risk the larger investment decision sitting behind it.