LandingAI’s DPT-2 in 2026: Why Agentic Document Extraction Finally Makes Sense

2026-04-17 · Source: Towards AI - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Intermediate, long

Summary

LandingAI has released its Agentic Document Extraction (ADE) system, powered by the DPT-2 (Document Pre-trained Transformer-2) model, with a Python SDK (`landingai-ade`) launched on March 11, 2026. ADE addresses the challenge of extracting structured data from visually complex documents like contracts, lab reports, and invoices, which often defeat traditional OCR and generic LLM approaches by discarding layout and visual context. DPT-2, announced September 30, 2025, processes both visual and textual elements jointly, introducing features like agentic table captioning with cell-level grounding, expanded chunk ontology for elements like checkboxes and barcodes, and refined figure captioning. The system operates via a three-API pipeline (Parse, Split, Extract) and employs an agentic loop for self-verification and replanning, enhancing accuracy on complex tables and long documents. LandingAI reports 99.16% accuracy on a subset of the DocVQA benchmark.

Key takeaway

For CTOs or VP of Engineering evaluating document intelligence solutions, LandingAI's ADE with DPT-2 offers a robust, vision-first approach for complex document extraction. You should prioritize the Zero Data Retention (ZDR) mode and EU data residency for GDPR compliance, and secure a Business Associate Agreement (BAA) for HIPAA-regulated data. Consider the Snowflake Native App for maximum data isolation if cloud processing is a concern, but be aware of the proprietary model and lack of self-hosting options.

Key insights

LandingAI's ADE system uses a vision-first, agentic approach to extract structured data from complex documents.

Principles

• Documents are visual objects, not just text.
• Agentic loops improve complex parsing accuracy.
• One-size-fits-all models fail on diverse document types.

Method

ADE employs a Parse-Split-Extract pipeline, using DPT-2 to jointly process visual and textual elements, and an agentic loop to plan, decide, and self-verify extraction tasks.

In practice

• Use `landingai-ade` SDK for document parsing.
• Define Pydantic schemas for field extraction.
• Configure Zero Data Retention for sensitive data.

Topics

• Agentic Document Extraction
• DPT-2 Model Architecture
• Document AI
• Computer Vision
• Enterprise Compliance

Code references

• landing-ai/ade-python
• landing-ai/agentic-doc

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• How to Build Your Own (DIY) Document Parsing Agent from Scratch — Agentic systems enhance GenAI document parsing by addressing VLM limitations like token limits and repetition.
• Extracting BOMs from Electrical Drawings with AI: Azure OpenAI GPT-5.4 + Azure Document Intelligence — Decomposing complex visual tasks into smaller, hybrid AI stages significantly improves accuracy and efficiency.
• LiteParse: Local Document Parsing for AI Agents — LiteParse is an open-source, local-first parsing tool for spatial text extraction from diverse document types.
• OCR Processor - Mistral AI — Mistral AI's OCR API extracts structured text from diverse document types, preserving formatting and providing confidence scores.
• Why Your Agents Can’t Read Enterprise Documents — and How to Fix It — Accurate document processing is foundational for trustworthy agentic workflows, especially with complex enterprise documents.
• [R] IDP Leaderboard: Open benchmark for document AI across 16 VLMs, 9,000+ documents, 3 benchmark suites — An open benchmark evaluates 16 VLMs on document understanding, revealing tight performance margins among top models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards AI - Medium.