Multi-Disease AI Diagnosis System Using Minimal Code and LLM-Powered Explanations
Atulya Thakur · 2025
This paper presents a novel approach to multi-disease diagnosis using a lightweight neural network architecture combined with large language model (LLM) powered explanations. We demonstrate how modern AI assistants can enable rapid development of functional healthcare AI systems with minimal code implementation. Our system achieves 72% accuracy on synthetic multi-disease classification tasks (Diabetes, Parkinson's Disease, and Healthy classifications) while providing human-readable explanations for each diagnosis through an integrated LLM reasoning module. The complete system, implemented in under 500 lines of code, demonstrates end-to-end functionality from data generation to model deployment, showcasing the potential for democratizing AI healthcare development. Key contributions include: (1) a streamlined architecture combining neural networks with LLM explanations, (2) automatic symbolic rule extraction for high-confidence predictions, (3) a deployable model export system, and (4) transparent documentation of AI-assisted development methodologies.