Guardoc says Amazon Nova cut medical document errors 46% and audit fines 70%
AWS ML Blog · rss · 2026-07-28
Guardoc Health uses Amazon Nova models on Bedrock to process fragmented medical documents at scale for long-term care facilities.
Reported results
- 46% reduction in documentation errors
- 70% fewer audit fines
- $400K+ annual ROI for a single facility
- Over 1 million documents per day on peak days
How the pipeline works
- Amazon Textract extracts text, handwriting, tables, and layout from PDFs.
- Documents are chunked by clinical meaning, not arbitrary size.
- Embeddings from Amazon Titan Text Embeddings V2 are stored in DynamoDB per patient.
- A pre-filter narrows candidates by document type and recency.
- An in-memory k-NN step retrieves the most relevant chunks.
- Nova 2 Lite removes obvious non-matches.
- Nova Pro handles the final multimodal review of raw PDF bytes, including layout, handwriting, signatures, and stamps.
AWS says the design uses the cheapest model capable of each stage, while preserving page-level traceability for clinical decisions.
More from Multimodal
- OrcaDub 1.0 launches speech-to-speech video dubbing with 4.83 MOS — qinzytech · 2026-07-28
- Kinetic Abyss shows a 2D animation made with Stable Audio 3 and LTX 2.3 — Tadeo111 · 2026-07-28
- How to turn reference photos into a convincing impasto oil-paint workflow — Suspicious-Koala-570 · 2026-07-28
- A Midjourney style workflow uses GPT breakdowns to narrow hundreds of image references — beechinour · 2026-07-28
- Krea2 Detail Enhancer LoRA Removes AI Fog Blur for Crisp Images — linoy_tsaban · 2026-07-28
- Creating a 2000s Pop-Punk Music Video Using Runway and Suno — notiansans · 2026-07-28