The gap is not the model. It is the knowledge.
The knowledge care runs on lives in the protocols and in the heads of the few people in every team who quietly do the work better. A general model does not bring the two together, not even with your documents added. A knowledge system does.
Written by
Founder and CEO
Essay9 min read
With Inora, 44.9% of working time comes back. We measure that by work process and by team, never by individual employee. But first: where the gap is.
03:10
Take a nurse on the night shift in a nursing home. It is 03:10. They are the only nurse in the building, as they are at many locations in the evening, at night and at the weekend. Anyone in doubt at that hour calls whoever is on call, or decides alone.
An acute admission arrives from the hospital’s emergency department. The nurse’s organisation has an agreement for this, and a help card: which steps, in what order, and what the first report must contain.
That knowledge exists. It just lives in two places, and at 03:10 the nurse can reach neither.
The first is the quality handbook. The second is the head of the colleague who never needs the card, because they have done it this way for years. They are not on shift tonight.
This is not an edge case. In Inora’s first year, 45% of questions arrived in the evening, at night or at the weekend. At those hours there is no colleague to ask.
Today I wrote down Inora’s story for the boards of care organisations. It starts at night, because that is where you see where the gap is. This is the long version, with the reasoning that did not fit into the story.
Two places
The handbook is where an organisation writes down what it has agreed. But no one opens the quality handbook in the middle of a shift, and anyone who does cannot always trust what they find.
What is not in the handbook is in people’s heads. Every team has a few people who do the work a little better, though no one calls it that. They know which version applies, which agreement the team actually follows and whom to call at night. When they leave, what they know leaves with them. One care organisation set Inora a goal in its annual report on quality and safety for 2025: a lasting team memory, so that the team depends less on knowledge in individual people’s heads.
Both places are needed, and both fall short. The handbook belongs to everyone but is out of reach. The head is within reach, but only for the colleague who happens to be working beside them. Inora brings the two together.
What a general model does
The usual answer to this problem is to give a language model your documents. I understand why. A language model writes fluently, understands a question typed the way people talk, and is simply there at 03:10.
But a general model given your documents is still a general model. It starts from everything it has ever read. Where your documents are silent, it fills in from what it read elsewhere: another hospital’s procedure, in another country. A help card that exists only as an image is invisible to it, so it fills that gap too. It makes up what is missing, and it sounds just as sure when it is wrong.
Nor does it know who is asking: a care assistant at vocational level 2 or the district nurse. And it does not know which version applies. Give it a folder with two versions of the same protocol, and it has no way of knowing which one the organisation adopted, which one has expired and which source comes first.
So the gap is not in what the model can do. The model can do more than enough. The gap is in the knowledge: it is not there, or it is there three times, or it is in someone’s head.
How a knowledge system turns it round
The computer, the internet and the smartphone put knowledge within everyone’s reach. They did not make it fit. What a nurse needs at 03:10 is not a library of books but a library that answers their question, in their situation, from what their own organisation has set down.
A knowledge system turns the work round. At its centre is not the model but the knowledge, and the people who answer for it. Inora is a care organisation’s knowledge system: it holds what the organisation knows and has agreed, keeps it current and keeps it within reach of everyone, from the care assistant at the bedside to the board. That comes down to four things.
Every source has an owner, a version and a layer. Inora keeps a register in step with the organisation’s own document system. It answers only from the current version, stops using a document that has expired or been withdrawn, and answers from a new version within minutes, without anyone at Axiomatic having to act. Each of the organisation’s sources is marked in the answer, and one tap shows the passage it rests on, with its source, layer and version.
The organisation decides what comes first. Its sources are its own documents, the regional agreements it adopts and the protocols its work agreements point to. It decides which of them leads. Where two of them disagree, the answer follows the leading one and says in one sentence that the other differs. Inora never decides which one is right. A source from outside the organisation never reaches the care worker.
Every gap goes on a list. When the answer is not in the sources, Inora says so, says where it looked and names who owns the subject. It does not guess. The topic goes as a task to the owner of that area of knowledge, without the question and without the name of whoever asked. This is where the two places meet. What was in someone’s head is written down, goes through the organisation’s own route for documents, and from then on it is in the source, for everyone, at 03:10 too. That is why, for every function, we first sit down with one colleague in that function and write down what a good answer looks like. The colleague already knows. Now it is in the system too.
The line lives in the code. For a question about one client, Inora puts the organisation’s own protocol beside the professional and flags what is missing; the professional decides. Where Inora stops is not an instruction to the model, which a clever question could talk it out of. It is in the code, and for every new version we count the errors at the line in both directions: what was wrongly answered and what was wrongly refused.
Back to 03:10. The help card has been read, and its owner has confirmed the text. Inora gives the nurse the steps of their own organisation’s night card, with the passage one tap away, and the four things the first report must contain: the reason for admission, cognition, particulars and long-term care (Wlz) status. They decide what to do for this client.
The least important part
In all of this, the model is the least important part. It is also the only part that can be replaced.
That was a choice from the start. The plan of 13 June 2025 tied Inora to no single language model. The model does what it is good at: writing. 38% of messages to Inora ask it to do something rather than find something, and 25% are a report, handover, letter or email. A language model is strong at that, as long as what it writes rests on something.
That way the organisation gets the best of both: the reliability of its own sources, and the language of the model.
When a better model comes along, we test it first on the questions of the organisation’s own teams, and only then do we replace the old one. The sources, their owners and versions, the order, the list and the line all stay where they are. That is the part the organisation has built, and it stays the organisation’s. Every protocol, guideline and handbook remains the property of the organisation that wrote it. What an organisation puts into Inora makes Inora better for that organisation, never for another.
Anyone who tells a care organisation that the model makes the difference is selling it the one part that will never be its own.
Who closes the gap
You do not see a gap in the knowledge from an office. You see it on the ward, at night, in a team’s own words. That is why the people who build Inora bring it into the teams themselves, team by team. Every implementation starts with the quality handbook, gone through with the quality adviser: who owns each document, which version applies, what has expired. What recurs in one team we fix at its cause, in Inora itself, so that the next team never meets it.
We do not bill hours. We earn nothing from a gap that stays open.
What it returns, and how we measure it
Inora gives care workers back 44.9% of their working time on searching, reporting and handover: 215 minutes of a 480-minute shift. Per team of 20 that is 9.0 FTE of time. Across Dutch nursing, residential and home care, 480,000 employees, it is the working time of 215,000 people: more than the 126,800 people the sector expects to be short in 2035.
We measure it by work process and by team, against a baseline taken before the start. Every month the board receives a report summed across all teams: what staff asked about, which topics have no answer yet, which gaps were closed and how much time Inora gave back.
Never by individual employee. That is not a detail of the report. Measure individual employees, and a knowledge system becomes a control system. Then no one asks the question that shows what they do not know, and that is exactly the question that tells the organisation where its knowledge falls short. No one in the organisation reads the conversations, and Inora assesses no one.
The knowledge is already there
The gap is not the model. It is the knowledge. That knowledge already exists, in your protocols and in the heads of your best people. Our work is to have it in one place at 03:10, with the passage one tap away and an owner behind it.
Ahsan Fazal, founder and CEO, 21 September 2026
Postscript, 2 November 2026
Inora is not a medical device. Inora supports triage: the professional triages, scores and decides. Inora puts the organisation’s own protocol beside the professional, flags what is missing and records the source and version of every answer; it does not diagnose or decide on treatment. Its intended use keeps it outside the medical device rules (The professional decides).