The patient's words are not the book's words

A patient says her husband is bossy at home. The rubric is MIND > DICTATORIAL. A mother says her son was late to teethe. The book files it under DENTITION. The books say coryza where the patient says cold, and loquacity where the family says he won’t stop talking. Kent files a craving under desires: search for “craving” and you find a handful of rubrics, search “desires” and you find hundreds.

None of this is a fault in the books. They are precise, written in the medical English of their time, and their structure is why repertorisation works. But it means every case needs a translation: from what the patient actually said into the words the book uses. Learning to do that takes years, and it is one of the real skills of the profession.

What it costs, in doctors’ own words

Indian practitioners have argued about this in public, and the discussion is worth reading. Dr Gaurang Dave puts the time plainly:

"Suppose we see 100pt per day & repertrisation require 10 minutes. Then 1000 minutes."

Dr Pritesh Patel, in the same discussion, on the old manual way:

"In the earlier days we had to Repertorise manually on a Repertory Sheet … and it was really tedious … hence most of us disliked it or avoided it … But, now with the laptops and Hom Softwares … we have a very versatile tool."

And the other side, which we should print too, from Dr Sameer Upadhayay:

"I see almost 100 cases in a day and repertories almost all the cases, that doesn't take much time if you know what to find and from where."

He is right. For a doctor who knows the materia medica deeply, the translation is quick because it happens in their head. No tool gives you that knowledge, and any tool sold as a replacement for it deserves doubt.

The doctor’s half and the clerical half

Repertorisation is really two jobs that usually get talked about as one:

• The doctor’s half: deciding what is characteristic in this case, which symptoms belong in the totality, and judging the result against the materia medica and the patient. This is the real thinking. No software does it, and none should pretend to.
• The clerical half: finding the rubric that matches what the patient said, turning pages across books that file things differently, listing what you took, and carrying the chart to the next visit. It needs skill, but it is still lookup and copying.

Software should only take the second half. Not because doctors are slow, but because ten minutes of page-turning per case is ten minutes of a trained mind doing index work.

What this looks like from the inside

We build this part in Homoeoclinics, so here is what the problem really looks like:

• The word gap is real. “Bossy” shares no letters with DICTATORIAL. “Late to teethe” shares none with DENTITION. A search that only matches the letters you type will never find these. So we keep a hand-checked list that links everyday phrasing to the book’s words, one entry at a time.
• Even the books disagree. Kent, Boger and the Boger–Boenninghausen repertory write agg. and amel.. Knerr, the Therapeutic Pocket Book and Boericke spell the words out. A search that insists on one form quietly misses the other books.
• Indian case notes help a little. Dr Ajit Kulkarni notes that "Indians narrate their complaints as they are" — plain description, not elaborate metaphor. Plain words translate more easily. But easier isn’t solved.

In the app, you write the case the way you already do — c/o, k/c/o, < and >, one finding per line — and it is matched to rubrics across a rich set of classical repertories. This happens on our own servers. Your case text is never sent to an outside AI service.

Sometimes you will also see results from a ninth book of our own. It helps when one symptom is written in different words in different books, putting those entries side by side, and each one still tells you which book it came from, so you can check the original.

Which half should AI take?

The same split is the simplest way to think about AI. Translating between languages, transcribing a consultation, tidying dictation, summarising a long history: these are clerical jobs, and AI is good at language. Choosing which rubric fits this patient is the doctor’s job, and handing it over changes what the software is for.

There is a practical reason to be careful, too. In a 2026 study, four AI chatbots were given 100 cases. When the same case was asked again on the same chatbot a few minutes later, the top remedy changed half the time.Healthcare 2026;14(7):909 Even a clerk who filed the same letter in a different drawer every morning wouldn’t keep the job.

So our search works from words first, and the word list can be edited. Its one smart helper runs on our own servers, cannot make up a rubric, and only steps in when your own words found nothing. More on this in why we don’t put an AI in the prescribing seat.

You check the list. You don’t obey it.

One rule holds all of this together. What comes back is a list of rubrics, shown above the remedies, where you can search, add or remove any of them. The remedies below are reference for you to check before prescribing. The software suggests wording; you decide the totality. A hard case is hard because of the doctor’s half of the work, and a better index doesn’t change that. What it changes is how much clerical work is left on your desk.