Semantic search in homeopathy

Semantic search represents both your query and the repertory or materia-medica corpus as numerical vectors, then ranks candidates by how close those vectors sit in a high-dimensional space. Where keyword search asks "which entries contain these exact words?", semantic search asks "which entries mean roughly the same thing?" — so a phrase the patient actually used, such as "she feels everyone is against her", surfaces the rubric MIND; DELUSIONS; persecuted, he is even though the two share almost no literal vocabulary.

How it works mechanically

A homeopathic semantic-search engine runs in four stages.

The decisive fact is that the embedding model, not the repertory, defines nearness. The model is trained on general-language text, so its notion of "close" reflects everyday usage, not the specialised vocabulary of Kent or Boenninghausen. This is the source of semantic search's signature failure mode — the plausible-but-wrong neighbour — and the reason the re-ranking stage exists. A tool that ships embedding plus nearest-neighbour search but skips re-ranking returns a candidate list that is fast and frequently misleading.

Semantic versus keyword search

The two modes answer different questions, and the disciplined prescriber uses both.

Reach for semantic search when the case notebook holds the patient's own words; reach for keyword search when confirming a rubric path you already have in mind. A defensive habit is to run the same case through both modes and compare the top candidates. Where the two lists agree, the rubric is anchored both in the surface vocabulary of the repertory and in the embedding model's sense of meaning. Where they disagree, that is a flagged item to read carefully before committing it to the chart. This is the posture Boenninghausen and Kent already encode — never trust a single rubric in isolation; verify against modalities and concomitants.

Search inside the materia medica

Semantic search is not confined to the repertory. The same technique ranks whole remedies by how strongly their materia-medica descriptions match a set of symptom phrases — a different clinical task from finding a single rubric. Here the corpus question sharpens. A tool that has embedded only abbreviated keynote lists returns shallow matches. One that has embedded full classical texts — Kent, Boericke, Hering, Clarke, Allen — ranks remedies on the texture of a proving rather than on a one-line summary.

Serious materia-medica search also exposes graded strictness: words in sequence, words in the same sentence, words in the same paragraph, and pure semantic match. The ability to choose strictness is itself an audit surface — a prescriber who reports a remedy match can state which mode produced it, and a peer can reproduce it.

Practical implementation

In current software, semantic search in the repertory accepts a free-text symptom — "fear of the dark", "headache better lying down" — and returns ranked rubrics from the available repertories, including translated Complete Repertory editions where the relevant translation is loaded. In the materia medica, it accepts one or more symptom phrases as separate entries and ranks remedies by match strength across selected classical sources, with a full-text-with-highlights view that jumps directly to the matched passages.

Two operational points are worth naming. Semantic-search stacks commonly rely on third-party embedding and vector-index providers (Google, Pinecone and similar); a serious clinical implementation contracts for zero-retention processing, refuses model-training reuse of submitted text, and operates under Business Associate Agreements where covered patient data is in play. Because a semantic-search query can carry patient language verbatim, the data posture is a clinical consideration, not a marketing footnote — read the provider's terms before pasting a case into the box. You can search the free repertory by symptom this way without paid access.

Where it sits in the canon

Semantic search is a retrieval interface — a clerical accelerant that shortens the distance between a patient's phrasing and the rubric vocabulary that names it. 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 describe a 200-year observational practice in which the rubric, the modality, and the constitutional reading carry the case. Semantic retrieval feeds that practice; it does not replace any step of it.

Verdict

Semantic search earns its place in case-taking when the implementation is backed by a real classical corpus, ships a re-ranking stage rather than raw nearest-neighbour output, surfaces match scores and source attribution, and exposes keyword search alongside it so you can cross-check — and when you treat the ranked list as input to judgement, not as a verdict. Companion notes on AI case analysis, natural-language homeopathic search, and AI rubric mapping sit alongside this one.

References

Hahnemann, S. Organon of Medicine, sixth edition.

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.

Boericke, W. Pocket Manual of Homoeopathic Materia Medica.

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

Allen, T. F. The Encyclopedia of Pure Materia Medica.

Vithoulkas, G. The Science of Homeopathy.

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