How Medi-Chi reasons, what it draws on, and how we treat your clinical data.
Medi-Chi is operated by Derek Doran, a registered Chinese medicine practitioner in Australia, trading as Medi-Chi and Eastern Medical Acupuncture. There is no anonymous holding company behind it.
Domain WHOIS is privacy-protected (standard registrar default), so these details are published here rather than in the registry record. Data handling, subprocessors, retention and breach response are set out on our Trust & Security page, and every change to the AI system — model, rules, retrieval libraries and citation re-checks — is logged on AI changes & verification.
Medi-Chi is built by a registered practitioner for other practitioners and students. In real clinical work, trust comes from accuracy and the ability to verify — not from a black box. This page sets out how our reasoning engine synthesises classical TCM theory with modern diagnostic standards, and what we do (and don't do) with the information you enter.
Medi-Chi isn't a general-purpose chatbot. Every case runs through a structured, two-layer analytical pipeline before any output is shown.
Presenting patterns are mapped through the primary TCM diagnostic systems: Zàng-Fǔ organ pattern differentiation, the Six Channels of the Shāng Hán Lùn, the Four Levels and Four Aspects of the Wēn Bìng school, Eight Principles, Five Elements, and San Jiao differentiation. The tool also cross-checks pulse, tongue, and constitutional signs against these frameworks.
Where the practitioner supplies Western clinical indicators — pathology results, physical exam findings, imaging notes, medications — Medi-Chi cross-references them against the TCM pattern shortlist, flags relevant red flags, and links out to verified PubMed-indexed citations rather than free-text speculation.
Our engine does not guess. It follows established TCM diagnostic pathways to map patient symptomatology to identified patterns, and each suggestion is accompanied by the reasoning trail and source framework so you can audit it against your own clinical judgement.
Practitioners and reviewers reasonably ask which AI actually runs under the hood. Here it is, plainly.
The evidence badges on our homepage are measured, not estimated. Here is the denominator, the method, and the honest breakdown as at 20 August 2026.
We take every unique PubMed ID cited across published Medi-Chi articles — 208 records at the last audit — and ask the PubMed E-utilities API for each record's publication type. Nothing is hand-classified, and a paper cited in five articles is counted once.
So 38% of the corpus sits in the highest tier (systematic review, meta-analysis or controlled trial), and 63% is either a trial or some form of review. 39% was published within the last five years, and the median publication year is 2018.
An earlier version of our homepage badge showed 82% for RCT/systematic-review coverage. That number came from a partial sample rather than the full corpus, and it was too high. When we re-ran the audit against all 208 cited records the true figure was 38%, and the badge has been corrected to match. We'd rather publish the smaller true number than defend a flattering one.
A scheduled job runs every Monday (03:30 UTC) and re-checks every cited PMID against PubMed. For each record it confirms the identifier still resolves, that the declared first author and publication year match the canonical record, and that PubMed has not tagged it Retracted Publication or Retraction of Publication. The same check runs as a build gate on every change to our article data, so a dead or mismatched citation cannot ship. Validation means identifier, metadata and retraction status — it is not a re-appraisal of each study's clinical quality.
When the weekly sweep flags a retraction, an automated pipeline searches PubMed for candidate replacements on the same topic, filtered to non-retracted trials, systematic reviews and meta-analyses. Each candidate is re-verified, then an AI step proposes the best match and rewrites the surrounding sentence so the claim stays true to the new source. The result is opened as a pull request with a full diff — nothing is published without a human merging it. “Auto-replaced” describes the detection and drafting, not unattended publishing.
At the 20 August 2026 audit, 0 of the 208 cited records were flagged as retracted. Every replacement that has run is listed on AI changes & verification.
PubMed publication types are applied by indexers and are imperfect — a well-conducted trial can be indexed only as “Journal Article”, which pushes it into our “other” bucket. Retraction notices can also lag the underlying concern by months. Study design is a proxy for quality, not a measure of it, and a high-tier design in a small acupuncture trial still carries real risk of bias. Read the source.
Reproduce these numbers yourself with scripts/audit-citation-corpus.mjs in our codebase; it calls the public PubMed API and prints the table above.
We do not use your patient input, clinical notes, or case data to train our underlying models. Your clinical insights remain yours alone. Prompts sent to third-party inference providers are transmitted for the single purpose of returning your result and are not retained for model improvement.
Medi-Chi aligns with the Chinese Medicine Board of Australia (AHPRA) advertising and privacy guidelines, the Australian Privacy Principles under the Privacy Act 1988, and HIPAA-style access controls including row-level security, encrypted transport (TLS 1.3), encryption at rest (AES-256), and automatic PHI redaction in audit logs. We are not an electronic medical record and do not present ourselves as one.
The system is designed to operate on the clinical markers you provide. It does not require identifiable personal information — name, date of birth, Medicare number, or address — to produce a diagnostic impression. Practitioners can run cases using pseudonymised patient references.
For full details, see our Privacy Policy.
Medi-Chi is actively maintained. The current system version and last updated date are shown at the top of this page and are advanced whenever the reasoning pipeline, pattern library, or citation base changes.
Diagnostic algorithms and pattern outputs are periodically reviewed by experienced TCM practitioners to ensure AI-generated patterns remain consistent with classical teaching and contemporary clinical practice. Corrections identified through this process are folded back into the pattern library rather than into training data.
Being straight about the gaps matters as much as listing the strengths. As of 20 August 2026, Medi-Chi has not:
Where any of that changes, it will be recorded here with the date and system version.
Medi-Chi is a decision support tool. It is intended as an adjunct to the qualified practitioner's clinical reasoning — a second set of structured eyes across pattern differentiation, point selection, and herbal considerations. It is not a substitute for a physical examination, pulse and tongue diagnosis performed in person, or the practitioner's own clinical judgement.
The final diagnostic and treatment decision rests solely with the licensed professional. Medi-Chi does not diagnose, treat, cure, or prevent disease, and its output is not medical advice to the general public.
If you spot a pattern mapping, citation, or safety flag that doesn't square with your clinical experience, we want to hear about it. Get in touch via the contact form and flag it as a methodology query — these go straight to the practitioner review queue.
For the intended purpose of each individual function — the diagnosis engine, tongue analysis, reverse diagnosis, reports and records — and what each one explicitly does not do, see our Regulatory Position.
Medi-Chi · Registered practitioner Derek Doran, AHPRA CMR0002211465 · ABN 26 600 839 506