Customers
Two deployments we are allowed to write about
Organisations that put us next to their own documents don't generally want that advertised. These two agreed to the story without the name. We'd rather show you two real ones than a wall of logos we can't stand behind.
Regulated European medtech company · anonymised
A million clinical files, searchable by question.
The company's value sits in its library: over a million clinical studies and procedures, accumulated over years. On paper it is an asset. In practice it behaved like a warehouse with no index. Everything was in there, and finding the right thing meant a specialist reading their way toward it.
Sector
Medtech, EU
Library
Over 1,000,000 clinical studies and procedures
Who uses it
Specialists, in their day-to-day research
Outcome
Days down to minutes
What was in the way
Why it had not been solved already.
-
01
The search was the job
Specialists were spending days digging through the library before the real work could start. That time was invisible on every plan and every budget. -
02
Scale broke the usual tools
Past a million documents, filename search and folder structure no longer work. The document you need exists, but nothing tells you which one it is. -
03
The answer had to be checkable
In clinical material, a confident summary with no source behind it is worse than no answer at all. Anything generated has to point at the file it came from.
What we built
What we built, in three parts.
-
01
Ingest the whole library
The full corpus of studies and procedures, processed and indexed, so a question runs against everything rather than against the folder someone remembered. -
02
Answers that point at the file
Every response names the exact document it came from, so a specialist verifies in seconds rather than trusting a paraphrase. -
03
Inside their own environment
The library runs inside their own environment, with no third-party AI vendor in the path, which is what allows a company holding clinical material to use AI on it at all.
What changed
What changed once it was running.
-
01
Days down to minutes
The outcome the client stated, and the one that pays for the project. -
02
The library became usable
A million documents that had been sitting in storage turned into something the organisation could question directly. -
03
Verification stayed cheap
Because the exact file comes back with the answer, checking is a click, not a second search.
“Over a million clinical studies and procedures, and specialists spending days digging through them. We built an AI that answers their questions and points to the exact file. Days became minutes.”
Children's therapy network · anonymised
Every therapist can see the whole patient file.
A therapy network runs on two things: the record of the child in front of you, and the accumulated knowledge of the organisation, meaning studies, worked therapy examples and specialist books. Both existed. Neither was reachable from the place the work gets done.
Sector
Children's therapy
Scope
The entire internal process, digitised
Who uses it
Therapists, during their sessions and preparation
Boundary
The clinical decision stays with the therapist
What was in the way
Why it had not been solved already.
-
01
The record and the knowledge were separate
A therapist could have the patient file or the reference library, but not both in the same place at the same time. -
02
Health data ruled out public AI
Special-category data under GDPR. Any tool that would have helped was also a tool they were not allowed to use. -
03
The judgement had to stay human
Nothing about this could be allowed to look like the software deciding treatment.
What we built
What we built, in three parts.
-
01
The whole process, digitised
The internal process itself moved online, rather than a tool being bolted alongside the old way of working. That is why the record is open in the same place as the question. -
02
The organisation's own library, answerable
Studies, therapy examples and specialist books, indexed so a therapist can ask them directly instead of remembering which book it was in. -
03
Cited, always
Every answer carries the source, which is what lets a therapist weigh it rather than take it on faith.
What changed
What changed once it was running.
-
01
The full file is open
Therapists work with the complete patient record rather than fragments held in different systems. -
02
The library answers back
Institutional knowledge that used to live with whoever had read it is now available to everyone in the network. -
03
The call is still theirs
The platform retrieves and cites. The clinical decision remains the therapist's, and the product is built that way rather than disclaiming its way there.
“We digitised their entire internal process. Therapists open the full patient record online and ask the organisation's own library: studies, therapy examples, specialist books. Every answer is cited. What to do next is still the therapist's call.”
Why there are no screenshots here
Both references are anonymised, and the material behind them is clinical. We don't publish the clients' names, their documents or their interfaces. What we can do is walk you through the same platform on your own files, and where the client agrees, arrange a reference call.
The shape they share
Three things both projects had in common. Yours will too.
-
01
The library became searchable by question
A million clinical studies in one case and a shelf of specialist books in the other. In both, material that had been sitting in storage became something staff could question directly. -
02
Every answer names its source
In clinical work an unsourced summary is worthless. Both deployments return the exact file or passage, so verification costs a click. -
03
No third-party AI vendor
Health data is special-category under GDPR. Running inside their own environment, with no third-party AI vendor ever seeing the data, is what made the project permissible at all.
Why the list is short.
A platform that exists because organisations don't want their documents leaving the building attracts clients who don't want their name leaving it either. Most deployments are covered by agreements that rule out a public write-up.
So this page will stay small on purpose. What we can offer instead is more useful than a logo wall. We will run the platform live on documents you bring, and where the client agrees, arrange a call with someone already using it.
A demo on your files
30 minutes, your documents, your questions. The fastest way to tell whether this is useful to you.
A reference call
Where the client agrees, we'll put you in touch with someone running it today.
The security review
Bring your DPO and your infrastructure team. The architecture is the product.
A scoped pilot
Live in your environment in under five days, on one team's library, before you commit further.
The third deployment could be yours
30-minute demo, on your files, no commitment.