Descriptions:
DeepLearning.AI has launched “Document AI: From OCR to Agentic Doc Extraction,” a new course developed in partnership with Landing AI. Taught by David Park and Andrea Drop—with Andrew Ng serving as executive chairman of Landing AI—the course addresses a pervasive bottleneck in enterprise AI: critical information locked inside PDFs, scanned reports, and complex documents that traditional extraction tools handle poorly.
The course opens by diagnosing why conventional OCR fails at scale: it strips away document structure, misreads multi-column layouts, drops table context, and ignores charts entirely. Learners first build an agentic extraction pipeline from scratch using layout detection combined with LLM-based reasoning, then advance to Landing AI’s Agentic Document Extraction (ADE) tool—which treats document pages as images, uses custom visual models to interpret tables and figures as spatial objects, and grounds every extracted field to its precise on-page location.
Beyond raw extraction, the course covers integrating ADE output into RAG pipelines and deploying a production-grade version on AWS using an event-driven architecture that automatically triggers document processing whenever new files arrive. Target applications include financial document analysis, medical record processing, and academic research review—use cases where structured, context-preserving extraction is essential for downstream AI systems to reason correctly.
📺 Source: DeepLearningAI · Published January 14, 2026
🏷️ Format: Course Lesson
