ai differential diagnosis
this is an experimental ai, for educational purposes only. do not use for real cases and do not rely on ai outputs for assessment or management of real patients.
diffyd --help
name
diffyd — ai-assisted differential diagnosis generator
usage
diffyd [--manifest] <vignette>
description

generates differential diagnoses from clinical vignettes using multiple ai perspectives. reduces anchoring bias and surfaces diagnoses that might otherwise be missed.

why

diagnostic error is a leading cause of patient harm. 10-15% of diagnoses are incorrect, often due to cognitive biases like anchoring.

diffyd explores whether ai can help clinicians think more broadly, especially for complex or atypical presentations.

pipeline
1. initial — baseline differential
2. perspectives — parallel ai review
· probabilist — most likely
· alarmist — dangerous / cant miss
· skeptic — alternatives
· anchoring breaker — cognitive biases
3. deduplication — merge duplicates
4. enrichment — likelihood, severity, features
output
· icd-10 code
· likelihood (highly likely → unlikely)
· severity (critical → mild)
· consistent / inconsistent features
· consensus score
limitations

ai models hallucinate. icd-10 codes may be wrong. not validated for clinical use. experimental prototype for education only.

contact
[back]