An AI-powered knowledge hub with citation-backed retrieval across seven repositories

The stakes

The clinicians, researchers, and policymakers the Koita Foundation supports operate from knowledge held in several places simultaneously, distributed across publications, standards repositories, expert directories, and institutional websites. Locating a particular clinical standard, or identifying the appropriate expert on a question, has required working through several systems manually with no reliable means of establishing whether all relevant material had been found. Those undertaking this work are the same people whose time the institution can least afford to allocate to searching.

The gap

Keyword search returns documents, leaving the user to find the answer inside them, and provides no assistance in assessing which source to rely upon or how two pieces of information relate to one another. Institutional knowledge assets were consequently underused, expertise was accessible only to those who already knew where to look, and considerable qualified time was allocated to manual research that produced no record of its own thoroughness.

What we engineered and operated

We engineered a knowledge intelligence platform that unifies seven or more repositories behind a single natural-language interface, combining retrieval-augmented generation with a knowledge graph. Deep crawling and automated indexing bring each connected source into the corpus; classification and summarisation are applied at ingest, and administrators retain control over which sources are admitted in the first instance. A user poses a question in ordinary language and receives a response assembled across every connected source simultaneously, with the underlying experts, institutions, publications, and standards explicitly mapped. Automated re-indexing is performed weekly, so every query reflects the current state of the sources.

How it held up under scrutiny

A response of this kind is usable only where its provenance is visible, so traceability was treated as a core requirement from the outset. Every response carries inline citations, a confidence score, and source attribution, and approximately 95% of responses can be traced to their origin, which enables a clinician or a policymaker to verify the evidence for themselves.