Systems & integration
Turning a national cultural institution into a connected digital exhibition
Six interactive visitor applications running across more than a hundred devices — touch tables, kiosks and screens — all fed and updated from one central content system, so curators could change an exhibit without anyone touching a device.
The challenge
The institution wanted its collection to be explorable rather than only viewable: visitors handling material interactively rather than reading a label beside a case.
The engineering problem was not any single application. It was that the finished installation would be well over a hundred devices spread across a large building, each showing different content, and all of it needing to stay current without someone walking round with a memory stick.
The context
Different exhibits needed genuinely different interactions. A currency collection wants close inspection of individual objects. A photographic archive wants browsing. Historic correspondence wants to be handled on a large surface by several people at once. Audio wants a completely different interface again.
Building one generic application for all of that would have served none of it well. Building six unrelated applications would have created six separate content-update problems.
The approach
I split the work along the seam between experience and content. Each exhibit type got its own application, designed for its own interaction and its own hardware. Every one of them was a client of the same content system.
That single decision is what made the installation maintainable at that scale, and it is the same decision I would make today for a fleet of devices in the field.
The solution
Six applications were built: detailed 3D exploration of a currency collection, an interactive photographic archive with drill-down on individual images, video, a large-surface application for exploring historic correspondence on touch tables, an interactive presentation of national figures with curated galleries, and an audio experience.
Behind them, a central management system holds all content and device registration. Content assigned to a device is delivered to it automatically — no manual intervention, no site visit, no per-device configuration. Curators change an exhibit centrally and the devices follow.
AI & automation
None. This predates that work, and the problem did not call for it — the requirement was reliable content distribution and rich interaction, both of which are deterministic problems.
It is included here because it is the same architectural problem I now solve for AI deployments: many endpoints, one source of truth, and updates that propagate without touching the endpoints individually.
Engineering
Content synchronisation to a large device fleet was the core problem: knowing what each device should have, detecting what it actually has, and reconciling the two without saturating the network or interrupting a visitor mid-interaction.
The applications had to be resilient in a way ordinary software does not — running unattended all day, recovering from restarts without a person present, and never showing an error screen to the public. 3D content had to perform on fixed installation hardware that could not be upgraded, which put real constraints on asset size and rendering.
Outcome
More than a hundred interactive devices ran as one connected exhibition, managed centrally, with content that curators could update themselves.
Visitors could handle the collection instead of only looking at it.
Client, employer and product names are deliberately withheld. Happy to talk through the specifics, and the architecture, on a call.
More work
Other engagements
AI & retrieval
A knowledge assistant that answers from your own documents
Documents turned into something people can simply ask — hybrid search rather than vector similarity alone, a reranker that cuts forty candidates to six, and a numbered citation on every answer pointing at the document and page it came from.
AI agents & real-time
A voice agent that takes the call, looks it up, and answers
A caller rings, speaks normally, and gets an answer drawn from live business data — transcription running while they are still talking, a model working out what they actually want, and a query against the systems that already hold the answer.
Automation & generative AI
An unattended production pipeline: a topic goes in, a finished video comes out
A pipeline that takes a topic, researches and scripts it, generates the narration and the visuals, renders the video and files the result — with no human step in the middle. Built partly as a product and partly to learn where multi-stage generative systems fail in practice.