Breakthroughs in machine learning indicate great promise for enhancing patient care, but difficulties in dealing with the complexity of systems and confidentiality in data prevent its applications in healthcare settings. This paper introduces IntegraMed, an INtelligent and Transparent Explainable system for Guided Research and Analysis in MEDicine. The system features an interpretable Adaptive Multi-Scale Attention network in conjunction with a locally deployed quantized language model, thus allowing clinicians to carry out complex analysis on tabular data in natural language without compromising data confidentiality. Usability testing with 50 users attained a 97% task success rate, an 82% System Usability Scale score, and favorable User Experience Questionnaire results. IntegraMed demonstrates how interpretable, locally executed AI can provide secure, accessible, and effective support for medical data analysis.