DiffyD — Differential Diagnosis

The questions to answer before every diagnosis.

De-identified case input. For clinical decision support only. AI can be wrong — verify all outputs against clinical evidence and your own judgement before acting.

How it works

A real case. The questions it deserves.

Paste a de-identified case, add a working diagnosis if you have one, and DiffyD generates the questions a senior colleague would push back with — alongside a full differential and the investigations that discriminate.

Community-acquired pneumonia62F · 3 weeks progressive breathlessness, dry cough, bilateral opacities, afebrile, CRP 45, WCC normal
Your consultant asks

If you’re leading with Community-acquired pneumonia

If this is pneumonia, why is she afebrile with a normal white cell count?

Typical CAP produces a leucocytosis and fever in most cases. Absence of both should expand your differential significantly.

What else could bilateral opacities in a 62-year-old ex-smoker mean?

Heart failure, malignancy with lymphangitis, and pulmonary fibrosis all produce this pattern.

Are you sure this isn't cardiac — has BNP been checked?

Bilateral alveolar opacities with dry cough in this age group make cardiogenic oedema the leading diagnosis until proven otherwise.

1Paste a case — any presentation, any complexity
2Get challenged on your anchor with a broader differential
3Click any diagnosis to generate targeted challenge questions

Why it works

When you’re stuck, the problem isn’t what you know. It’s what you’re looking for.

Complex and atypical cases are hard in a specific way. The moment you settle on a working diagnosis, it shapes every question you ask after it — what you examine, what you test for, what you discount. That’s not a failure of training. It’s how clinical thinking works. DiffyD interrupts that process before you commit: a broader differential, the questions a senior colleague would push back with, and the investigations that discriminate rather than confirm.

AnchoringIf you've committed to a diagnosis, we challenge it — the questions designed to expose what that lens is hiding.
Premature closureIf one answer fits, we keep looking. The second diagnosis doesn't announce itself — the differential does.
Knowledge gapsIf your differential is narrow, we broaden it. You can't consider what you haven't encountered. DiffyD has.

Paste a case. Get challenged.

Works on any presentation. Takes 30 seconds.

AI safety

What DiffyD can and can’t do

AI-assisted clinical tools introduce a specific category of risk that doesn’t exist with textbooks or colleagues. We’re building DiffyD with those risks in mind, and we think you should understand them before relying on this in practice.

Non-determinismSubmit the same case twice and DiffyD may generate different challenge questions. This isn't a malfunction — it's how large language models work, producing responses from a probability distribution rather than a fixed lookup. Treat every output as one perspective from a well-read but fallible colleague, not a reproducible investigation result.
BiasThe model learns from medical literature, which reflects the populations, presentations, and clinical priorities that have historically received the most research attention. Atypical presentations in underrepresented groups, rare conditions, and non-Western clinical contexts may be handled less reliably. Apply additional scrutiny when the case doesn't fit the standard mould.
How we're building thisDiffyD is in active development. We evaluate outputs against real clinical cases, refine our prompts when questions miss the mark, and are building structured evaluation frameworks before expanding into higher-stakes settings. Your feedback — especially when DiffyD gets something wrong — directly shapes the product.

Questions or concerns about how DiffyD handles a specific case? Write to us at hello@diffyd.com.