Towards Empowering the Offline Clinician: A Method for Enhancing Dermatology Reference Material Utility through Mobile Edge AI-Based Retrieval-Augmented Generation

Background: In 2024, we demonstrated the feasibility of using retrieval-augmented generation (RAG) on local laptops to create intelligent personal clinical tutors for dermatology education. However, the rapidly evolving capabilities of mobile hardware now allow for similar localised deployments on smartphones. Objective: This study investigates the feasibility of replicating a local, offline RAG system on a contemporary mid-range smartphone to provide a highly portable and secure clinical dermatology educational tool for clinicians and students. Methodology: Using a Redmi Note 15 5G (12 GB RAM, Snapdragon 6 Gen 3 processor), we deployed the 'Off Grid - Private AI Chat' app (Android). A 4-billion parameter model, Qwen 3.5 4B (Vision), was loaded on-device. A compressed PDF of a 234-page dermatology textbook was indexed into a local RAG knowledge base. All operations--model execution, indexing, and query processing--were executed natively and exclusively within the app without network connectivity. Results: The system successfully generated accurate, contextually relevant answers derived from the textbook. The model listed specific source text chunks used for each response, allowing for easy verification. Comparative analysis revealed that the RAG-enabled system prevented critical clinical inaccuracies observed in unconstrained Qwen 3.5 4B output by strictly grounding generation in the verified knowledge base. We provide technical guidance to facilitate high-performance execution on mid-range and flagship smartphones. Conclusion: The experiment confirms that mid-range mobile hardware is now sufficiently advanced to support localised, offline AI dermatology tutoring. This democratises access to sophisticated study aids while maintaining absolute data privacy, copyright protection, and offline flexibility.