SkinMind: Equipping Agentic RAG with Human-Inspired Graph Probing for Reliable Dermatology AI
SkinMind raised average scores by 8.70 and 8.89 points over direct answering across six dermatology visual question-answering benchmarks.
The gains came with GPT-5.4-mini and Gemini-2.5-Flash, respectively; the authors also report consistently outperforming competing RAG methods. Eight graph-probing tools let the system follow relationships and retrieve text or images iteratively, adjusting its search as it gathers evidence. Its knowledge graph draws on 62 million tokens from dermatology guidelines and textbooks plus 423,000 images, with more than one million nodes and two million edges.
HF Daily Papers' note
The gains came with GPT-5.4-mini and Gemini-2.5-Flash, respectively; the authors also report consistently outperforming competing RAG methods. Eight graph-probing tools let the system follow relationships and retrieve text or images iteratively, adjusting its search as it gathers evidence. Its knowledge graph draws on 62 million tokens from dermatology guidelines and textbooks plus 423,000 images, with more than one million nodes and two million edges.
HF Daily Papers' note
score 4