Bail decisions constitute a large share of criminal proceedings in Indian courts, yet they often suffer from delays, inconsistency, and limited transparency. More than 75% of India’s prison population consists of undertrial prisoners, highlighting the urgency of improving the efficiency of bail adjudication. In this work, we present the Indian Bail Prediction System (IBPS), an AI-assisted framework that predicts bail outcomes and generates legally grounded explanations from factual case attributes and statutory context. We construct a large-scale dataset of 89,969 bail judgments from 5 High Courts of India, enriched with structured annotations such as statutes, custody duration, criminal history, health conditions, and judicial reasoning. Using this dataset, we design a two-stage architecture that combines a transformer-based classifier for outcome prediction with a fine-tuned large language model for rationale generation. Our results demonstrate the feasibility of combining structured legal data and large language models to develop interpretable AI systems that support data-driven analysis of bail decisions.