AI case analysis in homeopathy

AI case analysis takes the raw material of a consultation — typed notes, a photographed handwritten page, a recorded conversation — and returns a structured set of repertory rubrics, symptom themes, and remedy hypotheses for the prescriber to test against the case. It does not prescribe. It accelerates the clerical work between hearing a symptom and finding the rubric that names it, leaving the interpretive work — the constitutional reading, the modality cross-check, the materia-medica confirmation — where the tradition has always placed it.

The four stages of an AI case-analysis pipeline

A serious pipeline has four mechanical stages, and a prescriber who understands all four can audit the output rather than trust it blindly.

Intake is the stage most prescribers underweight. An OCR step that silently misreads "worse 4 pm" as "worse 4 am" corrupts every downstream stage, and a transcription step that drops the half-sentence in which the patient described their core delusion removes exactly the strange, rare, and peculiar material on which a Kentian or Sankaran reading turns. The output of any AI case analysis is only as honest as its intake — the audit trail of what text the model actually saw matters more than the polish of the final rubric list.

Symptom extraction and the strange, rare, and peculiar

Extraction is where the technique either earns its place or fails the method. Hahnemann's Organon, aphorisms 153 and following, directs the prescriber to weight the striking, singular, uncommon, and peculiar signs of the case above the common symptoms that many remedies share. A naive extractor inverts that priority: it lifts the loud, frequent symptoms ("anxiety", "headache") and buries the quiet, individualising detail that actually narrows the remedy field.

A well-designed extractor pulls symptoms in a deliberate priority order — strange, rare, and peculiar first, then modalities, sensations, locations, causations, and concomitants — and avoids overly generic rubrics in favour of rubrics carrying roughly 15 to 100 remedies. That design is a direct encoding of aphorism 153: it pushes the pipeline toward the rubrics that discriminate between remedies rather than the rubrics that everything covers. Ask the same question of every vendor — does the extractor reach for the peculiar or for the obvious? — because the answer determines whether the candidate set is clinically usable or merely plausible.

Rubric mapping and the corpus that backs it

Once symptoms are extracted, each is mapped to repertory rubrics by semantic retrieval: the symptom phrase is embedded into a vector, compared against pre-embedded rubric paths, and the nearest candidates are returned. The quality of this stage depends almost entirely on the corpus. A pipeline that has indexed Kent's Repertory, Boenninghausen's Therapeutic Pocket Book, and a current Complete Repertory edition with rubric paths intact returns candidates a classical prescriber can act on. A pipeline indexed against paraphrased modern summaries returns candidates that read well and prescribe poorly. The mechanics of this retrieval layer are examined in depth in the companion piece on AI rubric mapping.

Two surfaces separate a defensible mapping from an opaque one. First, the rubric path shown verbatim — for example, MIND; FEAR; dark, of — so the prescriber reads the rubric in its repertory context before adding it to a chart. Second, a visible match score and source attribution, so a strong match against an exact Kent rubric is distinguishable from a weak match against a paraphrased textbook line. Without both, the prescriber inherits a rubric list with no audit trail to defend on peer review.

Manual versus AI-assisted case analysis

The honest comparison is not "AI versus the prescriber" — it is the same prescriber working with two different toolsets.

The last row is the load-bearing one. No AI step transfers clinical responsibility away from the prescriber; the candidate set is an input to judgement, exactly as a repertory chart has always been. Vithoulkas and Sankaran both describe case analysis as an act of interpretation that integrates the totality of the case, not a counting exercise — a faster way to assemble the count does not displace the interpretation.

Working surfaces in practice

The technique typically appears in a handful of concrete surfaces. A notes-to-rubrics tool analyses typed or pasted consultation notes and suggests rubrics, with an option to analyse only a highlighted selection when the notes exceed the character limit. An upload-notes feature runs OCR over photographed handwriting and inserts the recognised text for analysis. A live mode turns typed symptoms, one per line, into auto-added rubrics during a live consultation, with a convention that lines ending in two dots are treated as informational and not analysed. A live audio mode records a consultation, transcribes it, extracts strange-rare-and-peculiar symptoms, and auto-adds matching rubrics, in person or online. A case-timeline feature can generate dated case events from notes by the same extraction step.

Two implementation details are prescribing-relevant rather than marketing footnotes. Cost should be explicit — credit-gated or seat-gated, not buried — so the per-case economics are visible before adoption. And because every AI step is a covered-data event under most clinic privacy regimes, the vendor's data posture matters: zero-retention handling under a Business Associate Agreement, no reuse of submitted patient data for model training, and — for audio specifically — no persisted recording, with only the transcript and summary written back to the case. Prescribers who want to test the extraction quality on their own notebook can work directly inside the free repertory and case tools at similia.io and judge the rubric shortlist on real cases.

Where it sits inside the tradition

AI case analysis is a retrieval-and-extraction interface. It does not change the prescribing judgement — it accelerates the assembly of candidate rubrics so judgement can operate on a broader, better-organised set of inputs. The canon — Hahnemann's Organon, Kent's Repertory and Lectures, Hering's Guiding Symptoms, Boenninghausen's Therapeutic Pocket Book, Boger's Synoptic Key, Vithoulkas's Science of Homeopathy, and Sankaran's sensation-method volumes — describes a 200-year observational practice in which the rubric, the modality, and the constitutional reading are the load-bearing artefacts. Within that practice, AI case analysis is a clerical accelerant for an interpretive method.

The technique earns its place in a prescriber's workflow when the tool weights strange-rare-and-peculiar symptoms in extraction, surfaces rubric paths and source attribution verbatim, and treats patient data with a published zero-retention posture — and when the prescriber treats the candidate set as an input to judgement, not a substitute for it. The companion methodology on semantic search in homeopathy and the broader AI in homeopathy overview cover adjacent ground.

References

Hahnemann, S. Organon of Medicine, sixth edition — aphorisms 153 and following on the striking and peculiar symptoms as the basis of remedy selection.

Kent, J. T. Repertory of the Homoeopathic Materia Medica; Lectures on Homoeopathic Materia Medica.

Hering, C. Guiding Symptoms of Our Materia Medica.

von Boenninghausen, C. Therapeutic Pocket Book.

Boger, C. M. Synoptic Key of the Materia Medica.

Clarke, J. H. A Dictionary of Practical Materia Medica.

Boericke, W. Pocket Manual of Homoeopathic Materia Medica.

Vithoulkas, G. The Science of Homeopathy.

Sankaran, R. The Substance of Homoeopathy and the sensation-method volumes.