ADE DPT-2: A Major Step Forward in Document Intelligence
Last week, we introduced Document Pre-trained Transformer 2 (DPT-2) — the latest upgrade powering Agentic Document Extraction (ADE).
Here’s a quick look at what’s new, and why it matters for real-world document processing.
ADE DPT-2 combines structured deep learning models with agentic workflows to handle some of the toughest document challenges with higher accuracy and trust. Complex tables, skewed scans, and embedded visual elements like signatures, stamps, and QR codes are now parsed with unprecedented fidelity and grounding.
Andrew Ng walks through real-world examples of ADE DPT-2 in action — from parsing 1,000+ cell tables to extracting patient results from lab reports — so you can see exactly how this upgrade changes what’s possible.
Core Features of ADE DPT-2
Recommended by LinkedIn
Availability
ADE DPT-2 is now in preview. You can try it directly in the ADE Playground or access it via the ADE APIs.
In addition, ADE’s Parse and Extract APIs have been decoupled, making it easier to parse documents once and then run multiple field extraction passes without reprocessing. Together with recent speed gains, ADE can now process hundreds to thousands of pages per minute while maintaining layout awareness and visual grounding.
For teams in finance, healthcare, insurance, and compliance, these updates reduce costly manual review while maintaining accuracy, traceability, and auditability.
📖 Industry Spotlight: In a recent Forbes feature, Victor Dey highlights how ADE DPT-2 is redefining document intelligence for finance, healthcare, and compliance — bringing structure and trust to complex, visual data. Read the full article here
We can’t wait to see what you build with ADE DPT-2.
— The LandingAI Team
https://www.linkedin.com/pulse/information-theory-everything-unified-understanding-anurag-dongare-tm8we?trackingId=02FZiC7BQCuKgOXD8GSh%2FA%3D%3D&lipi=urn%3Ali%3Apage%3Ad_flagship3_profile_view_base%3Babj4J3OPRBG5WYH5MiILyA%3D%3D
Thomas Nokin
AI will enable so much in just the next 1 year that neither can it be described today what it will enable, nor, after 1 year, would today's stage of AI development seem anything like even an "incipient stage"