Chronic Kidney Disease remains a problem for global healthcare solutions, with worldwide affliction rate approximately 10%. Early detection is essential in CKD because if undetected, it might lead to the development of end-stage renal disease (ESRD). There has been significant research done in CKD prediction using machine learning (ML). However, there has been only limited real-world implementation due to the challenges of unstructured and noisy electronic health records (EHRs). To overcome these limitations, we present a novel hybrid pipeline that combines document processing, with ML-driven prediction. The method integrates computer vision using Optical Character Recognition (OCR) for preprocessing unlabeled EHRs and intelligent feature engineering to mine clinically significant biomarkers. Then CKD classification is done based on a hybrid model. We create our model with Random Forest, XGBoost and a Neural Network (NN) for ensemble learning. One of the main innovations of this framework is the inclusion of confidence metrics, improving interpretability and clinical decision-making. The proposed pipeline shows the potential to increase efficiency in a clinical setting, as well as real-time diagnostics with high confidence, which helps improve trust in AI-powered tools and solutions in healthcare.