Ask Particle
A conversational layer over institutional art archives, built with Particle:
provenance, restoration records and curatorial notes made queryable in plain language.
Museums hold far more than what they publish: provenance files, restoration records, curatorial correspondence, internal cataloguing. That knowledge is fragmented, unstructured, and invisible to anyone trying to ask a serious question about a work.
We built the technical infrastructure end to end — ingestion, normalisation, RAG pipelines, and an attribution system that traces every answer back to the institution that made it possible.
Source-traced answers returned in conversation, across web and mobile.
The hard part was never the model. It was making a claim traceable: which archive, which record, which institution — attached to every sentence the system produces.
That constraint shaped the architecture. Attribution is not bolted on at the end; it is the reason the institutions agreed to open the files at all.
Ingestion, normalisation, RAG pipelines, fine-tuning, and the usage accounting that turns a query into a monthly figure for the institution that answered it.