Guardoc Health introduced a clinical documentation pipeline that processes more than one million documents daily using Amazon Nova models through Bedrock. The system combines text extraction, retrieval and multimodal reasoning to interpret complex clinical records containing printed text, handwriting, signatures, stamps and irregular layouts. Amazon Textract first extracts text and structure before records are chunked along clinical boundaries and embedded with Amazon Titan Text Embeddings V2. Patient-partitioned retrieval in DynamoDB, document-type and recency filters, k-nearest-neighbour search and a Nova 2 Lite pass narrow the candidate pages before Amazon Nova Pro analyzes the original PDFs for difficult cases. For long-term care providers, Guardoc says the approach has reduced documentation errors, audit fines and compliance costs while accelerating corrections. The architecture highlights how cost-tiered retrieval-augmented generation and targeted multimodal AI can support large-scale clinical documentation workflows.