Radiology Report Generation (RRG) aims to reduce the burden of clinical documentation, yet generating reports that remain visually grounded in the input image remains challenging. Current medical vision-language models (Med-VLMs) often rely on curated preference data, retrieval augmentation, or large-scale labeled corpora, but remain prone to hallucinated findings due to weak cross-modal alignment. To address this challenge, we propose GRAIL: Grounded Radiology Alignment of Medical Vision-Language Models for Radiology Report Generation, a two-stage multimodal alignment framework for grounded RRG. First, Clinically Informed Textual Reasoning improves report structure and medical terminology using verifiable language rewards with periodic clinical guidance. Second, Expert-guided Visual Reasoning leverages a domain expert to rank generated reports using image-report alignment scores, encouraging visually grounded report generation without curated preference pairs or retrieval data. Experiments on multiple benchmarks show that GRAIL consistently improves both report generation quality and clinical accuracy, achieving relative gains over strong post-training baselines by up to 15% in clinical accuracy and 13% in generation quality. GRAIL also outperforms proprietary general-purpose and medical VLMs, highlighting the effectiveness of self-supervised multimodal post-alignment for reducing hallucinations in visually grounded radiology report generation.

Enterprises are increasingly adopting Generative AI applications to extract insights from large volumes of multimodal documents in domains such as finance, law, healthcare, and industry. These documents contain structured and unstructured data (images, charts, handwritten texts, etc.) requiring robust AI systems for effective retrieval and comprehension. Recent advancements in Retrieval-Augmented Generation (RAG) frameworks and Vision-Language Models (VLMs) have improved retrieval performance on multimodal documents by processing pages as images. However, large-scale deployment remains challenging due to the high cost of LLM API usage and the slower inference speed of image-based processing of pages compared to text-based processing. To address these challenges, we propose EcoDoc, a cost-effective multimodal document processing system that dynamically selects the processing modalities for each page as an image or text based on page characteristics and query intent. Our experimental evaluation on TAT-DQA and DocVQA benchmarks shows that EcoDoc reduces average query processing latency by up to 2.29× and cost by up to 10×, without compromising accuracy.