WhatDx lets a busy clinician use frontier AI without spending the consultation engineering prompts. The patient completes the deep history, record upload and iterative questioning at home; the clinician gets an English research-fellow dossier built around the specific question that needs solving.
WhatDx is designed to do work that is valuable precisely because it is slow: reconstructing chronology, asking detailed follow-up questions, connecting organ systems, researching uncommon explanations and showing what evidence would actually discriminate between them.
Adaptive questioning can explore onset, sequence, family history, exposures, treatment response, prior labels and contradictions over many rounds instead of compressing years of illness into a few minutes.
When the presentation justifies it, WhatDx can research inherited and rare-disease families that fit the age, family pattern, laboratory clues or multisystem phenotype, while keeping them as hypotheses for clinician evaluation rather than automated diagnoses.
Neurological, gastrointestinal, rheumatological, endocrine, dermatological and other clues can be assessed on one timeline, which may surface a shared mechanism that fragmented care does not naturally reveal.
WhatDx can ask whether apparent treatment failure reflects the wrong diagnosis, the wrong mechanism, inadequate exposure, absorption, adherence, interactions or a competing process rather than simply recommending another treatment.
The investigation is asked to identify the next best move for the patient's desired outcome: which evidence would change the strategy, what can be deferred, and what realistic fallback exists if the ideal test or referral is not accessible.
A motivated patient or caregiver can organize the case, answer detailed questions and return with a structured dossier. That can make follow-up more focused and help patients feel that a difficult case is still moving forward.
The technology is most useful when the patient or caregiver can reliably participate in a detailed asynchronous investigation.
In the clinician-suggested pathway the patient is explicitly asked what you want WhatDx to investigate. The best objective names the uncertainty that would change management or understanding.
A useful objective preserves room for the existing diagnosis to be correct, incomplete or wrong. If you already have a favored hypothesis, include the reason it matters but ask WhatDx to challenge it rather than confirm it.
The report stays evidence-linked and is intended for clinician review, not autonomous diagnosis or prescribing.
The clinician-suggested report is intentionally layered. The first screen is for rapid clinical orientation; the deeper sections preserve the reasoning, evidence and patient interview behind it.
The objective, most important synthesis, leading branches, largest uncertainty and next best move appear before the long narrative.
Each meaningful branch shows its rationale, supporting patient-specific features, inconsistencies, the best discriminating evidence and what the next move becomes if the branch is supported.
A compact transcript summary shows what WhatDx asked and what the patient actually answered, so the clinician can distinguish patient history from model interpretation.
What can be done now, what requires new evidence, what can be deferred and which fallback remains if the technically ideal path is inaccessible.
Low-probability possibilities are grouped by mechanism or phenotype. A named rare or genetic disorder is promoted only when the presentation gives a specific reason to do so.
When appropriate, a clinician-supervised therapeutic trial may be framed as an evidence-generating strategy. Registered clinical trials are surfaced only after current registry verification and never as an eligibility determination.
These are fictional composites created to demonstrate the workflow. They are not real WhatDx outcomes and should not be used as medical advice.
A patient has years of episodic abdominal pain and swelling treated repeatedly as allergy. The interview identifies a family pattern and episodes without urticaria.
Research-fellow value: groups hereditary bradykinin-mediated disorders as a high-value hypothesis family, shows what fits and what does not, and identifies the discriminating evidence rather than listing dozens of rare diseases.
The working label blames a medication. A longer chronology shows sensory symptoms began before that exposure and gastrointestinal symptoms predated the neuropathy.
Research-fellow value: exposes the timeline mismatch, reframes the causal question and prioritizes old laboratory trends, absorption/nutritional clues, inflammatory context and selected hereditary considerations.
Each specialty has a plausible local explanation, but no one view accounts for episodic neuropathic symptoms, bowel disturbance, inflammatory markers and intermittent skin findings.
Research-fellow value: builds a cross-system strategy tree, ranks shared-mechanism explanations, separates common reversible causes from grouped rare possibilities and tells the clinician which next evidence would change the branch.
The digital referral card sends the patient to the clinician-suggested WhatDx pathway. It explains that WhatDx is independent, that the patient pays the normal research-access fee directly to WhatDx, and that the resulting dossier is for clinician review.
WhatDx does not pay clinicians for referrals and has no commercial relationship with hospitals, laboratories, pharmaceutical companies or medical providers whose products or services might appear in a report.
