About · Philosophy
How medlogicai.org thinks
A platform of clinical decision-support modules where the algorithm structure is borrowed from foundational ENT textbooks, the content of every node is freshly synthesised from current peer-reviewed literature by AI, and nothing reaches a clinician without a named human reviewer signing off.
How it works
Four steps from textbook structure to live content
01
Skeleton
The structure of each decision algorithm — anatomical section, node ids and branch logic — is derived from foundational ENT textbooks. The textbook is inspiration for the structure only; no clinical content is copied. Every node's body, threshold, statistics and citations are empty after the skeleton stage.
02
Synthesis
Claude synthesises each node's content from current peer-reviewed literature — PubMed, CrossRef and Europe PMC — with full provenance captured (sources used, model version, prompt version, generation timestamp). Every clinical claim cites a PubMed source by PMID. Every statistic is pinned to a citation.
03
Review
Nothing reaches the public site without a named clinician reviewing and approving the proposed content through the reviewer dashboard. Actions are Approve / Reject / Edit-and-Approve, each recorded to the audit log with the reviewer's identity. There is no auto-merge tier — a one-character typo fix and a full node rewrite go through the same gate.
04
Live
Approved content is versioned and goes live with an explicit "Last reviewed by [name] on [date]" footer. AI-generated illustrations carry a visible disclosure badge with the reviewer's name. Prior versions are retained. The same pipeline drives nightly refresh — when the evidence moves, the content moves with it.
Standards we hold
What we will not compromise on
Every clinical claim cites a PubMed source.
No prose without a PMID. The contract is enforced at schema level via chapter_quality_contract: v1 and re-validated on every build.
Every statistic is pinned to evidence.
Each `pinned_stat` carries its source citation and a one-sentence evidence note. No floating numbers, no “studies show”.
Reviewers are identifiable, named clinicians.
Every approved change is signed by a real reviewer whose identity is recorded in the audit log. The reviewer's name appears on the live page footer alongside the date of approval.
AI-generated illustrations carry a visible disclosure.
Every rendered AI image shows a badge — “AI-generated illustration, reviewed by [clinician] on [date]” — and carries the same provenance trail as text content.
No PHI is collected, transmitted, or stored.
The PHI scanner runs on every diff. The platform is decision support, not a patient record; nothing about an identifiable patient ever enters this system.
Every change has an audit-chain hash.
Content versions, reviewer actions, and AI assistant interactions are all recorded in a tamper-evident hash chain. Any break in the chain blocks deploys.
Audience
For whom — and not for whom
Built for
- Practising ENT clinicians who want current, evidence-anchored decision support at the point of care.
- ENT, audiology, neurology and otolaryngology trainees building pattern recognition and a working evidence base.
- Vestibular therapists, audiologists and other allied professionals reading vestibular and auditory workups.
Not for
- Patients seeking medical advice. This site is not patient-facing and is not a substitute for consulting a qualified clinician.
- Clinical decisions made without independent judgement. The platform is decision support, not a regulated medical device. Clinicians remain completely responsible for interpretation, differential, and management.
- Storing or transmitting any patient-identifiable information. No PHI is collected by design.
The decisions behind this
Read the architecture decisions
Every load-bearing design choice on this platform is recorded as an Architecture Decision Record (ADR). The page you are reading distils these for a clinician audience; the ADRs themselves are the authoritative source.
ADR 0002
Platform-first architecture
Why every disease module is built on a shared platform layer with strict module isolation enforced by CI.
ADR 0004
Portal multi-zone architecture
How medlogicai.org runs as a portal with each module on its own subdomain, deployed independently but discoverable from one place.
ADR 0005 · the editorial gate
Literature-synthesised content, reviewer-gated, AI-generated illustrations with disclosure
The decision that makes this site what it is: textbook as inspiration for structure only, every node freshly synthesised from current peer-reviewed literature, every change gated through a named human reviewer, every AI image carrying a visible disclosure badge. The four-step pipeline above is its implementation.
Feedback
Spotted an outdated citation, a node where the evidence has moved, or a reviewer claim worth challenging? Write in. The reviewer team reads every message and changes ship faster when an external clinician flags something.