Venture
Diversity GPT
Realistic clinician training through mixture-of-experts personalities.
The problem
Clinician training needs conversational practice that feels real - not generic chatbot replies that miss tone, uncertainty, and clinical personality.
Our approach
A mixture-of-experts router sends each turn to specialist models tuned for different clinician personalities and conversational styles. Shared dialogue state carries tone, uncertainty, and case context across turns, so practice sessions behave like distinct clinicians rather than one flat assistant.
Overview
Diversity GPT was built for high-fidelity clinical conversation practice. The MoE setup keeps personalities separable and responses situationally consistent, which makes training scenarios more transferable to real encounters. Early development benefited from YC mentorship and input from a Stanford Medicine PhD professor.
Capabilities
What it does
Mixture-of-experts routing across clinician personality specialists
Dialogue state for tone, uncertainty, and case continuity
Scenario packs for realistic training conversations
Evaluation focused on personality fidelity, not just answer correctness
Stack