Turning years of scattered documentation into a searchable knowledge base: a research org's story
A research organization used a custom AI knowledge base — built on retrieval over their own documents, not general web knowledge — to unify years of scattered findings into one searchable system, cutting time spent searching for existing research by 6 hours per person per week.
The problem: institutional knowledge nobody could actually find
The organization had years of genuinely valuable research behind it — but that research lived in a scatter of shared drives, individual inboxes, old project folders, and the memories of whoever happened to work on a given study. New staff had no reliable way to find out what the organization already knew. Experienced staff spent hours re-deriving findings that existed somewhere, if only they could remember where.
This is a common failure mode for research-heavy organizations: the work itself is rigorous, but the system for retrieving it is whatever informal habits grew up over the years. Nobody designed it. It just accumulated.
Why a shared drive and search bar weren't enough
Their first instinct, like most teams, was better folder structure and a faster search tool. That helps at the margins, but it doesn't solve the real problem: keyword search only works if you already know roughly what you're looking for and what it was called. Research findings rarely get remembered by their exact filename — people remember the topic, the client, or the rough conclusion, not the string of text that would match a keyword search.
What they actually needed was a system that could answer a question like 'what have we found about X in the last three years' and pull the relevant answer from across every document that touched it — not a faster way to browse folders.
What we built
We built a knowledge base system using retrieval-augmented generation over the organization's own documents — reports, past findings, internal notes — so answers are grounded in what the organization has actually produced, not general knowledge from the open internet.
Staff can ask a plain-language question and get an answer synthesized from the relevant documents, with the source material cited so they can verify it and read further. It sits wherever staff already work, rather than asking people to learn or remember to open a separate tool.
Keeping it accurate as new research comes in
A knowledge base that goes stale is worse than no knowledge base — it teaches people not to trust it. New documents are indexed into the system as they're produced, so the answer to 'what do we know about X' stays current rather than freezing at the day the tool launched.
We also built in source citations on every answer by design, so staff aren't asked to take the system's word for it — they can trace any answer back to the original document in seconds.
The result: 6 hours a week back per person
Staff now spend roughly 6 fewer hours a week searching for research that already existed — time that goes back into producing new work instead of rediscovering old work. Just as important, junior staff can now find institutional knowledge on their own instead of depending on whoever happened to remember the answer.
The organization also has, for the first time, a genuine picture of what it collectively knows — which turned out to be useful well beyond day-to-day search, including spotting gaps in their own research coverage.
Is this a fit for your organization?
This pattern shows up anywhere valuable knowledge outlives the people who produced it: research bodies, consultancies, agencies, and any team where 'ask around until someone remembers' has become the default search strategy. The tell is usually the same complaint, repeated in different words — 'I know we've looked at this before, I just can't find it.'
If that sounds familiar, the fix is rarely 'reorganize the folders again.' It's building retrieval around how people actually ask questions. Worth a conversation before you spend more time re-doing work that's already been done once.
