AI Semantic Search in Homeopathic Repertory Practice – Similia

Simone Ruggeri
AI Research Scientist, Similia Co-Founder

Disclosure: The author is a technologist and AI researcher, not a licensed homeopathic practitioner. Clinical interpretations are based on published materia medica literature. The author is co-founder of Similia, a homeopathic software platform; this affiliation is disclosed for transparency. All search examples are sample inputs constructed for educational purposes and contain no patient-identifiable data.

Disclaimer: This article is for educational purposes only and does not constitute medical advice. This content should not delay or replace professional medical care. For diagnosis and treatment decisions, consult a qualified healthcare professional. Homeopathic treatment should be conducted under the guidance of a trained practitioner.

At a Glance
Semantic search is an AI technique that translates a practitioner’s natural-language symptom description into matching repertory rubrics across multiple sources — from classical repertories like Kent to modern compiled works like the Complete Repertory and Murphy’s MetaRepertory. It bridges the gap between how patients describe symptoms today and the specialized terminology each repertory uses. This article demonstrates cross-repertory results for sample queries, examines the representation bias that shapes which remedies surface in repertory analysis, and discusses the practical benefits and limitations of the technology.

Key Findings

• Multi-source rubric discovery: A single natural-language query can surface relevant rubrics from Kent, Murphy, Complete Repertory, and Suggesta simultaneously, eliminating the need to search each source independently.
• Chapter-boundary discovery: Semantic search returns rubrics across multiple repertory chapters (e.g., MIND and SLEEP, or SKIN and RESPIRATION) for a single query, catching entries that manual chapter-by-chapter search frequently misses.
• Multi-author materia medica comparison: Searching remedy profiles across authors reveals how different writers emphasize different facets of the same remedy, compressing what would otherwise require consulting several texts side by side.
• Broader remedy discovery: Polychrests dominate repertory analysis partly because a self-reinforcing cycle — more prescribing leads to more clinical confirmation, which leads to more rubric entries — inflates their presence relative to less-mapped remedies. Semantic search on the materia medica compresses the differentiation step: for example, Kent’s SLEEP – WAKING – 3 a.m. lists nineteen remedies, but only materia medica research reveals that Nux vomica’s 3 a.m. waking centers on racing thoughts, Kali carbonicum’s on fear of death when alone, and Calcarea arsenicosa’s on perspiration. Pulling up those differentiating passages across multiple authors in minutes rather than hours makes every remedy in a rubric practically accessible — not just the familiar polychrests.
• Clinical judgment remains essential: Semantic similarity is not always clinical equivalence. Automated pipelines without practitioner oversight might produce unreliable rankings regardless of the search technology used.

Why Is Repertory Language So Difficult to Search?
Vocabulary mismatch between clinical language and controlled medical vocabularies remains one of the most persistent barriers to effective health information retrieval (Wang et al., 2018). The same problem applies to homeopathic repertories, but with the added complexity of terminology spanning two centuries — from Kent’s 19th-century conventions to the modern compiled structure of the Complete Repertory — and hundreds of thousands of entries organized under different editorial philosophies.

Consider a patient who says: “My joints ache when it rains, and I feel stiff in the mornings, but once I get moving it eases up.” A practitioner must convert this into rubrics such as EXTREMITIES – PAIN – Joints – damp cold weather (Kent), extremities – pain – joints – weather – damp, wet, in (19) (Complete), or Joints – PAIN – damp weather, in (Murphy). Each repertory structures the same concept differently, and none use language resembling how the patient spoke.

Even experienced practitioners might miss rubrics because they searched for the wrong synonym or looked in the wrong chapter. Keyword-based search fails here because it depends on exact term matching — typing “joint pain wet weather” returns nothing if the repertory uses “damp cold weather” instead.

Semantic search addresses this gap by operating on meaning rather than exact words. Building on dense vector retrieval approaches (Mikolov et al., 2013; Karpukhin et al., 2020), the system converts a natural-language query into a dense vector and compares it against pre-computed embeddings of rubric paths, returning ranked results regardless of the specific words used.

How Does Semantic Search Bridge Repertory Language Across Eras?

Sample Results
The following examples are sample results from an AI semantic search system that indexes repertory databases spanning classical and modern sources. Rankings may vary with different configurations, corpus versions, or query phrasing. Each query was run against Kent’s Repertory, Murphy’s MetaRepertory, the Complete Repertory, and the Suggesta Repertory simultaneously.

Example 1: Burning Stomach Pain Better From Warm Drinks

Sample query: burning stomach pain better from warm drinks

This is how a practitioner might describe an Arsenicum album keynote in everyday language. The query combines a symptom quality (burning), a location (stomach), and a modality (amelioration from warm drinks) — none of which use standard repertory terminology. Here is what AI semantic search returned:

Rank	Source	Rubric Path
1	Suggesta	GASTROINTESTINAL – Stomach pain – burning pains – warm – amel. by warm drinks
2	Kent	STOMACH – PAIN – burning – warm drinks amel.
3	Complete	stomach – pain – burning – drinks – amel. – warm
4	Murphy	Stomach – PAIN – warm, drinks – amel.
5	Complete	stomach – pain – drinks – warm – amel.
6	Kent	STOMACH – PAIN – warm drinks amel.
7	Suggesta	GASTROINTESTINAL – Stomach pain – general – food and drinks – warm or hot – drinks – warm
8	Complete	stomach – pain – warmth amel.

Four repertories, four different ways of encoding the same clinical observation. Suggesta places it under GASTROINTESTINAL – Stomach pain – burning pains; Kent uses STOMACH – PAIN – burning; the Complete Repertory structures it as stomach – pain – burning – drinks – amel. – warm; Murphy compresses it to Stomach – PAIN – warm, drinks – amel. A practitioner searching any single repertory would find only one phrasing. The AI search also correctly excluded rubrics describing the opposite modality (stomach – pain – drinks – warm – agg., i.e. pain worse from warm drinks), which appeared in the raw candidate pool. This kind of multi-source discovery is something manual repertory lookup cannot achieve without consulting each source independently.

Example 2: Thirst for Cold Water

Sample query: thirst large quantities cold water

This query describes a well-known keynote of Phosphorus. The results revealed how different repertories categorize the same observation:

Rank	Source	Rubric Path
1	Suggesta	GENERALITIES – Food and drinks – cold – drinks, water – desires – large quantities of cold water
2	Murphy	Food Thirst – COLD, drinks, water – desires – large quantities of cold water
3	Complete	stomach – thirst – drinks, cold
4	Complete	stomach – thirst – large quantities, for
5	Complete	stomach – thirst – unquenchable, constant – sipping of cold water

Three repertories, three organizing principles. Suggesta treats thirst as a constitutional general. Murphy groups it in a dedicated Food Thirst chapter. The Complete Repertory anchors it anatomically under stomach – thirst. Searching only one repertory finds just one framing; semantic search surfaces all three simultaneously.

Example 3: Insomnia With Midnight Anxiety

Sample query: insomnia anxiety restlessness after midnight

This symptom spans the SLEEP and MIND chapters. Classical repertories separate these, and a practitioner must know to look in both. The semantic search bridged chapter boundaries automatically:

Rank	Source	Rubric Path
1	Complete	sleep – sleeplessness – midnight – after – anxiety, from
2	Complete	mind – restlessness, nervousness – midnight – after
3	Complete	mind – anxiety – midnight
4	Kent	MIND – ANXIETY – midnight,
5	Kent	MIND – ANXIETY – midnight, – after
6	Murphy	Mind Emotions – ANXIETY, general – midnight, general
7	Suggesta	SLEEP – restless – midnight – after

The top-ranked Complete Repertory rubric merges sleep disturbance with its emotional cause in a single entry. Kent separates these into distinct chapter entries. A practitioner searching only the MIND chapter would miss the combined SLEEP rubric — exactly the gap semantic search fills.

Example 4: Suppressed Eruption and Asthma

Sample query: skin eruption suppressed asthma

This query touches one of the foundational principles in homeopathic thinking: the relationship between suppressed skin conditions and deeper pathology. The results were striking — not for what they found, but for where they found it:

Rank	Source	Rubric Path
1	Complete	respiration – asthmatic – eruptions – suppressed, after
2	Complete	skin – eruptions – rash – itching – alternating with asthma
3	Complete	skin – eruptions – rash – tightness of chest alternating with asthma
4	Complete	skin – eruptions – psoriasis – alternating with asthma
5	Kent	SKIN – ERUPTIONS – alternating – with – asthma
6	Complete	skin – eruptions – alternating with – respiratory complaints – asthma
7	Complete	chest – eruptions – rash – alternating with asthma
8	Kent	RESPIRATION – ASTHMATIC – eruptions,after suppressed
9	Suggesta	SKIN – Skin illnesses – eruptions – concomitant – alternating with – asthma
10	Murphy	Lungs Breathing – ASTHMA, general – rash, after suppression of acute

One query, three body chapters (SKIN, RESPIRATION, CHEST), four repertories, and a clinically critical distinction that every practitioner should notice: Kent separates “suppressed” eruptions (a causative concept rooted in Hering’s direction of cure) from “alternating” patterns (a temporal relationship). These are different clinical observations, and the relevant rubrics live in different chapters. The Complete Repertory further distinguishes by eruption type — itching rash, psoriasis, general rash — offering granularity that a single-source search would never reveal. Murphy, meanwhile, uses “rash” where Kent uses “eruptions,” adding yet another vocabulary gap that semantic search bridges effortlessly.

A practitioner looking only in the RESPIRATION chapter of Kent would find ASTHMATIC – eruptions, after suppressed but miss the SKIN chapter entries about alternating patterns. One looking only under SKIN would miss the respiratory connection entirely. Semantic search makes this kind of cross-chapter, cross-repertory discovery routine rather than exceptional.

What Happens When You Search Materia Medica Across Multiple Authors?

Repertory search matches symptoms to rubrics. Materia medica search uses the same technology but applies it to a different corpus: sentences and paragraphs within remedy profiles written by different authors. The same natural-language query runs against all indexed texts simultaneously, returning passages that would otherwise require reading each author’s work independently.

Cross-Author Example: Grief and Disappointed Love

Sample query: “ailments from grief and disappointed love, sighing, sensation of lump in throat”

Most practitioners would expect Ignatia or Natrum muriaticum. The search ran across six sources: Murphy, Pitt (Thematic and Comparative), Mangialavori, Meditative Provings, and Griffith.

Source	Remedy	Theme/Passage Summary
Murphy (Nature’s MM 4th ed.)	Hyoscyamus	Ailments from grief, disappointed love
Mangialavori	Hyoscyamus	Humiliation and disappointed love; tries to attract attention through extreme behavior
Pitt (Comparative)	Ignatia	Ailments from grief, anger, fright; cannot weep; sighing, sobbing
Murphy	Ignatia	Sensation of lump in throat; sighing; silent grief
Murphy	Natrum muriaticum	Ailments from disappointed love; dwells on past events; cannot weep in public
Pitt (Comparative)	Natrum muriaticum	Grief held in; cannot weep before others; desires solitude to cry
Griffith	Staphysagria	Suppressed grief, indignation; swallows anger; feels humiliated
Meditative Provings	Ruby	Heart pain from grief; feeling of loss and abandonment

Hyoscyamus ranked first because its Murphy entry contains the exact phrase “ailments from grief, disappointed love” — a near-perfect textual match. This illustrates a key limitation: semantic similarity measures textual closeness, not clinical weight. No experienced practitioner would rank Hyoscyamus above Ignatia for this presentation. Ignatia’s text uses different phrasing (“cannot weep,” “sighing, sobbing”) that is clinically equivalent but semantically less similar to the query string.

What the search does accomplish is surfacing Staphysagria and Ruby alongside the expected remedies. Both are clinically relevant for grief — Staphysagria for suppressed indignation and swallowed anger, Ruby (from meditative provings) for heart pain from loss — but neither would appear in a conventional rubric-based search for the grief syndrome. The technology does not replace the practitioner’s knowledge that Ignatia is the leading remedy here; it extends the differential by surfacing remedies the practitioner might not have considered.

Does Semantic Search Replace Traditional Repertorization?

No. Semantic search is a lookup accelerator, not a replacement for the clinical reasoning that drives repertorization. It changes only the first step: finding the right rubrics more quickly and across more sources. The practitioner still selects rubrics, weights them by hierarchy (generals, particulars, mentals, keynotes), and analyzes remedy coverage across the totality of the case — the individualization process that ultimately points toward the simillimum.

How to Use Semantic Search Results in Practice

For practitioners incorporating semantic search into their workflow:

1. Query: Enter the symptom description in natural language, including modalities, location, and qualifying details.
2. Review candidates: Examine ranked rubrics across repertories. Note which sources surfaced each result and whether the rubric captures the intended symptom.
3. Open the full entry: Click through to the complete rubric — review the remedy list, grades, and sub-rubrics directly. This is equivalent to looking up the rubric in the paper repertory, but without needing to know the exact path in advance.
4. Select rubrics based on case totality: Choose rubrics representing the characteristic symptoms — generals, striking particulars, and mentals — rather than accepting every high-similarity result.
5. Repertorize with clinical judgment: Weight selected rubrics according to Hahnemannian hierarchy. The final prescription rests on the practitioner’s assessment of the totality, not on similarity scores.

How Do Different Repertories Structure the Same Symptom?

Running semantic searches across multiple repertories reveals how dramatically organizational structure varies for the same clinical concept. Here is a comparison for “irritability worse from contradiction”:

Source	Rubric Path	Phrasing Strategy
Suggesta	MIND – Irritability – contradiction, during	Temporal marker (“during”)
Murphy	Mind Emotions – IRRITABILITY, general – contradiction, from	Causal marker (“from”)
Complete	mind – irritability – contradiction, from	Matches Murphy’s causal framing
Suggesta	MIND – Irritability – contradiction, during – slightest	Adds degree qualifier
Complete	mind – irritability – contradiction, from – slightest	Adds degree qualifier

Suggesta uses “during” as a temporal marker, implying simultaneity. Murphy and Complete use “from,” implying causation. Both offer a sub-rubric for the “slightest” contradiction — subtle but clinically meaningful differences that single-repertory searches miss. Similarly, the query “desire for open air ameliorates” returned rubrics from the mind, generalities, and extremities chapters simultaneously, linked by the modality rather than the body part — a cross-chapter linking that manual search cannot achieve without checking each section independently.

Why Do Some Remedies Dominate Repertory Results?

A question that arises naturally from cross-repertory search is why the same remedies keep appearing at the top of repertory analyses. The answer involves a self-reinforcing cycle that has shaped repertory development since the earliest compilations.

The Confirmation-Representation Cycle

Polychrests — remedies like Sulphur, Lycopodium, Natrum muriaticum, and Pulsatilla — appear in thousands of rubrics across every major repertory. Their dominance is partly clinical: they address broad constitutional patterns. But it is also partly structural. The more a remedy is prescribed, the more opportunities clinicians have to observe and confirm its symptoms. Each confirmation adds weight to existing rubric entries or generates new ones. Over generations, this cycle inflates a polychrest’s rubric count relative to lesser-prescribed remedies that may address similar symptoms but have been less thoroughly mapped.

The result is that a smaller remedy can be underused not because it is intrinsically less effective for a given symptom picture, but because it appears in fewer rubrics, making it harder to reach through conventional repertorization. When a practitioner searches manually, the remedies with the most rubric entries naturally dominate the analysis — regardless of whether those entries reflect genuine clinical breadth or simply more extensive documentation.

The Proving-to-Repertory Pipeline

Jeremy Sherr has examined a related distortion from the opposite direction: the process by which provings are converted into repertory entries (Sherr, 2017). His analysis identifies several points where editorial decisions shape which symptoms reach the repertory and how prominently they appear. A prover conducting their first proving may include every observed detail; a professional repertoriser working without deep familiarity with the proving’s essential nature may give equal weight to incidental and characteristic symptoms. Once entries reach a repertory, they are rarely removed — there is no systematic mechanism for contesting an entry based on non-observation in clinical practice.

This means repertory representation reflects not only clinical reality but also the proving methodology, the editorial standards of the repertoriser, and the accumulated weight of historical usage. Two remedies with identical clinical relevance for a given symptom can have vastly different rubric counts simply because one was proved more extensively or repertorised more liberally.

What This Means for Semantic Search
Semantic search does not solve the representation problem, but it shifts the discovery dynamic in a specific way. Because it ranks by textual similarity rather than rubric count or prescribing frequency, it can surface remedies that conventional repertorization overlooks. In the materia medica examples above, querying “cannot weep” surfaced Gelsemium alongside Natrum muriaticum — not because Gelsemium is necessarily a better match for the full case, but because its text contained language semantically close to that specific concept. Splitting the case into four independent queries revealed that only Nat-m. covered all four concepts, while Gelsemium matched just one.

Consider a concrete example. A practitioner working a case of nocturnal anxiety opens Kent’s Repertory and looks up SLEEP – WAKING – 3 a.m. The rubric contains nineteen remedies: Nux vomica and Sulphur in bold (grade 3), Kali carbonicum and Coffea in italics (grade 2), and fifteen others at grade 1, including Arsenicum, China, Sepia, Graphites, and Zincum. If the practitioner also checks MIND – ANXIETY – midnight, – 3 a.m., only two remedies appear — Arsenicum (grade 3) and Silicea (grade 1). Kali arsenicicum and Calcarea arsenicosa, both well-known for nocturnal aggravation, are not listed in either rubric.

The repertory tells you which remedies belong in these rubrics and how strongly they are graded. It does not tell you how each remedy expresses the 3 a.m. symptom or what differentiates one from another constitutionally. For that, you need the materia medica — and with nineteen remedies to evaluate, the research load is substantial. Traditionally, this means opening Murphy, Clarke, Boericke, or Pitt for each remedy and reading through the full profile to find the relevant passages. In practice, many practitioners skip this step entirely, defaulting to the polychrest they already know — typically Arsenicum or Nux vomica.

Semantic materia medica search compresses this research. Querying “waking at 3am with anxiety and restlessness” across Murphy, Pitt (Thematic and Comparative), and other indexed sources returns the relevant passages for each remedy, ranked by relevance. The top results paint a remarkably differentiated picture:

• Kali arsenicicum — Pitt: “Wakes up fearful, anxious, and restless, 1–3am.” A second query on fear and heart anxiety surfaces: “Fears gradually take over, especially about the heart, that he/she will have a heart attack.” And Pitt’s thematic analysis adds: “Duty, work, family, order: conscientiously focused on these ideas.” The remedy combines the Kali sense of rigid duty with the Arsenicum nocturnal restlessness and specific cardiac anxiety.
• Kali carbonicum — Pitt: “Waking at 2–4am: Wakes with anxiety,” “Waking at 2–4am: Many fears arise then,” “Waking at 2–4am: Cannot sleep.” Murphy adds: “Fear of death when alone.” Pitt (Comparative): “Fearful, anxious, being alone, extremely anxious and insecure.” The 2–4 a.m. window is characteristic, and the fear of being alone differentiates Kali-c. from the cardiac focus of Kali-ar.
• Nux vomica — Murphy: “Awakens at 3 a.m., lies awake for hours, with a rush of thoughts.” Murphy also: “Cannot sleep after 3 a.m. until towards morning.” Pitt adds: “Sleepless from rush of ideas” and “Feels terrible on waking in morning.” Unlike the Kalis, whose 3 a.m. waking centers on anxiety and fear, Nux vomica’s presentation is dominated by mental overactivity — the mind that will not stop.
• Calcarea arsenicosa — Murphy: “After 3 a.m., sleepless, restless and perspiring.” Murphy also: “Night sweat after 3 a.m.” The perspiration modality — absent in the Kalis and Nux vomica — is the distinguishing feature, combining Calcarea’s tendency toward perspiration with the Arsenicum nocturnal timeline.
• Magnesia carbonica — Pitt: “Waking at 3am.” Murphy adds: “Awakes at 2 a.m. or 3 a.m. and cannot fall to sleep. Unrefreshing sleep, more tired in the morning than on retiring.” Where Nux vomica lies awake with racing thoughts, Mag-c. simply cannot return to sleep, and wakes more exhausted than before.

In five minutes and three queries, the practitioner has extracted the differentiating features that would take over an hour of reading full materia medica profiles to compile: Kali-ar.’s cardiac fears, Kali-c.’s fear of death when alone, Nux-v.’s rush of thoughts, Calc-ar.’s nocturnal perspiration, and Mag-c.’s unrefreshing exhaustion. Each passage links directly to its source text for verification and deeper study.

The repertory already contains these remedies; semantic search on the materia medica reveals how they differ. The practitioner still evaluates each result against the totality of the case, the remedy’s proving quality, and their own clinical experience. The technology compresses the research; clinical judgment determines what to do with it.

What Are the Practical Limitations of This Technology?

Semantic similarity is not clinical equivalence. A rubric can be semantically close to a query while being clinically irrelevant. In the grief search, Hyoscyamus outranked Ignatia because its text contained the exact matching phrase while Ignatia’s wording differed slightly.

Representation bias affects all repertory-based tools. As discussed above, polychrests are overrepresented in repertories due to a self-reinforcing cycle of prescribing, confirmation, and documentation. Semantic search does not correct for this bias — it bypasses rubric count as a ranking factor, which can surface underrepresented remedies but also returns results that lack the clinical validation depth of well-proved polychrests. Practitioners should consider a remedy’s proving history and clinical track record alongside its semantic match score.

Results depend on the source texts indexed. If a repertory edition is not indexed, its rubrics will not appear. Different repertories vary not only in terminology but in scope and provenance: Kent’s Repertory is a classical work with approximately 68,000 rubrics; the Complete Repertory is a modern compilation synthesizing multiple source repertories into over 400,000 rubrics. These are fundamentally different kinds of works, and results reflect the indexed corpus, not the universe of homeopathic knowledge.

The technology does not interpret remedy relationships. Clinical synthesis — which symptoms are characteristic versus common, how rubrics relate to one another, and how remedy relationships inform final selection — remains the practitioner’s responsibility.

When to consult a practitioner. Readers exploring semantic search tools for self-study should remember that this technology is not a substitute for clinical training. Homeopathic case analysis, remedy selection, and posology require professional judgment. If you are experiencing health concerns, consult a qualified healthcare professional or trained homeopathic practitioner before making treatment decisions.

Privacy considerations. Practitioners using AI-powered tools should verify that their platform encrypts data in transit and at rest, does not use patient data for model training, and complies with applicable regulations (HIPAA, GDPR). Queries should never contain personally identifiable patient information.

Methodology Note
The repertory examples were generated on 3 March 2026 using an AI-powered semantic search system that indexes rubric paths from Kent’s Repertory (6th ed.), Murphy’s MetaRepertory (3rd ed.), the Complete Repertory (2026 digital ed.), and the Suggesta Repertory. Kent’s Repertory is a classical work; Murphy’s MetaRepertory, the Complete Repertory, and Suggesta are modern compiled or authored repertories — the system indexes sources spanning the full history of homeopathic repertory development to maximize rubric coverage. Queries are converted into dense vector representations and matched against pre-computed embeddings of rubric paths following dense retrieval approaches described in References 10–11. The materia medica examples used the same approach against passages from Murphy, Pitt (Thematic and Comparative), Mangialavori, Meditative Provings, and Griffith. The exact embedding model version and hyperparameters are proprietary. Rankings depend on embedding configuration, corpus version, and query phrasing. Tables are truncated to top-ranked results. Repertory and materia medica content is indexed under license from the respective publishers. The platform used for these examples is Similia (similia.io).

Frequently Asked Questions

What is semantic search in homeopathy?
Semantic search is an AI technique that matches the meaning of a practitioner’s query against repertory rubrics and materia medica passages. Unlike keyword search, which requires exact term matching, it understands synonyms, paraphrases, and related concepts, bridging the gap between modern clinical language and the terminology used across both classical and modern homeopathic repertories.

How does AI semantic search produce its results?
The system converts a practitioner’s natural-language query into a mathematical representation of its meaning (a dense vector) and compares it against pre-computed representations of every rubric and materia medica passage in the database. Rubrics and passages whose meaning is closest to the query are ranked highest. The process scans millions of entries in milliseconds and returns a short list of the most relevant results across all indexed sources simultaneously.

Can semantic search replace learning repertory language?
No. Understanding repertory structure, rubric hierarchy, and the conventions of specific repertories remains essential. Semantic search may accelerate lookup and help discover rubrics a practitioner might miss, but it does not substitute for foundational knowledge of how repertories are organized.

Why do different repertories phrase the same symptom differently?
Each repertory reflects its author’s organizational philosophy and era. Kent uses anatomical hierarchy in the conventions of his time. Murphy groups by clinical category. The Complete Repertory — a modern compilation, not a classical work — synthesizes multiple source repertories with contemporary normalization. The same clinical observation can appear under different chapter headings with different modifier structures.

How is semantic search different from keyword search?
Keyword search requires typing the exact terms used in the repertory. If a rubric says “damp cold weather” but you search “wet weather,” keyword search may return nothing. Semantic search matches on meaning, so both phrases map to the same concept and return the relevant rubric.

References

1. Kent, J.T. Repertory of the Homoeopathic Materia Medica. 6th American Edition, 1957. B. Jain Publishers.
2. Murphy, R. Nature’s Materia Medica. 4th Edition, 2021. Lotus Health Institute.
3. Murphy, R. Homeopathic Clinical Repertory (MetaRepertory). 3rd Revised Edition, 2005. B. Jain Publishers.
4. van Zandvoort, R. Complete Repertory. 2026 Edition (digital). Roger van Zandvoort.
5. Mangialavori, M. Materia Medica Clinica. 1–3. Matrix Editrice.
6. Pitt, R. Comparative Materia Medica. Homeopathic Educational Services.
7. Pitt, R. Thematic Materia Medica. 2017. Homeopathic Educational Services.
8. Hahnemann, S. Organon of Medicine. 6th Edition. Translated by W. Boericke, 1921. B. Jain Publishers.
9. Griffith, C. The New Materia Medica. Watkins Publishing.
10. Mikolov, T. et al. “Efficient Estimation of Word Representations in Vector Space.” arXiv:1301.3781, 2013.
11. Karpukhin, V. et al. “Dense Passage Retrieval for Open-Domain Question Answering.” arXiv:2004.04906, 2020.
12. Wang, Y. et al. “Clinical Information Extraction Applications: A Literature Review.” Journal of Biomedical Informatics, 86, 57–68, 2018.
13. Sherr, J. “The Prover and the Repertoriser.” Hpathy.com, 2017.

Last updated: March 2026

About the author: Simone Ruggeri is an AI Research Scientist and co-founder of Similia (similia.io), a cloud-based homeopathy platform that indexes 14+ repertories — both classical and modern — alongside multiple materia medica sources for natural-language querying and AI-assisted clinical workflows. He focuses on the intersection of machine learning and homeopathic methodology. For background on the information retrieval techniques underlying this work, see References 10–11.

Marco Ruggeri
Founder: https://www.similia.io/
Email : info@similia.io