Guardoc Health uses Amazon Nova to process medical documents

Guardoc Health, a company specializing in medical data processing, has deployed a pipeline based on the Amazon Nova model family, available through Amazon Bedrock, to transform clinical documentation in long-term care facilities. The system addresses three core challenges: detecting specific medical conditions with high recall, reliably interpreting PDF checkboxes, and accurately extracting information from mixed-format documents. For disease classification, it uses Retrieval Augmented Generation (RAG): first, Amazon Textract extracts text, then the data is segmented into clinically meaningful chunks, vectorized using Amazon Titan Text Embeddings V2, and stored in Amazon DynamoDB. After pre-filtering by document type, recency, and patient context, a k-nearest neighbors (k-NN) search is performed, then Amazon Nova 2 Lite filters out obvious mismatches, and finally Amazon Nova Pro performs multimodal recognition on raw PDF bytes, returning the final classification with page-level attribution. For PDF processing, Guardoc Health uses Amazon Nova Pro, which has shown improved layout interpretation, checkbox status, and handwritten annotation recognition. Handwriting recognition is critical for physician attestation fields and patient-reported symptom sections. For medication data extraction, a hybrid pipeline is used: Amazon Textract performs optical character recognition (OCR) for clean tables, while Amazon Nova Pro handles complex cases such as handwritten additions and non-standard formats. This has reduced documentation errors by 46%, decreased audit penalties by 70%, and delivered an annual return on investment of over $400,000 per facility.