LandingAI — IDP Software Vendor

Visual AI company founded by Andrew Ng specializing in agentic document extraction and computer vision through specialized transformer models.

Overview

LandingAI was founded in 2017 by Andrew Ng in Mountain View, California. Ng previously co-founded Coursera, led Google Brain, and served as chief scientist at Baidu. He stepped down as CEO on August 24, 2024 to focus on AI Fund; Dan William Maloney now leads the company. LandingAI has staked out "agentic document extraction" as a distinct category among IDP software vendors, positioning its approach against both traditional OCR pipelines and general-purpose large language models (LLMs).

The September 2025 launch of Document Pre-trained Transformer-2 (DPT-2) is the clearest expression of that strategy: a model trained specifically on document structure, including gridline-free tables, merged cells, and non-standard layouts, rather than adapted from a general foundation model. The company claims 99.16% accuracy on DocVQA, a visual question-answering benchmark for real-world documents, though no competitor score on the same test has been published independently, making the figure a floor rather than a competitive position. An AIMultiple evaluation of 60 document images ranked LandingAI first at 69/100 composite, ahead of Mistral OCR, Claude 3.7 Sonnet, OpenAI o3-mini, and Docsumo. Competitor scores were not disclosed, so the margin of the lead is unknown.

The credibility picture has a concrete gap. Deep Analysis Senior Analyst Dan Lucarini explicitly stated, citing LandingAI's own webinar, that Agentic Document Extraction (ADE) does not yet have confidence scoring for data extraction. He calls this a requirement for any enterprise production deployment. LandingAI's own product page lists "confidence scoring and audit-ready traceability per extracted field" as a current capability. The discrepancy is unresolved: either the feature shipped between the webinar and the page update, or the marketing page is ahead of the actual product. Lucarini's open invitation for LandingAI to brief analysts had not been answered as of his publication date.

LandingAI was also absent from the ISG Buyers Guide for IDP Platforms published February 11, 2026, the same week it received editorial analyst coverage. This suggests it has not yet met the revenue, customer count, or analyst engagement thresholds for formal enterprise evaluation inclusion. Tracxn ranks LandingAI 7th among 216 active competitors with a score of 58/100, positioning it above early-stage startups but below mega-funded competitors like Primer ($237M), Hebbia ($161M), and Vectara ($73.5M). The company sits alongside Reducto AI (valued at $600M), Unstructured, and automat as Bay Area IDP startups collectively raising approximately $200M in a market that added 100 new vendor names in the past 12 months.

Strategic partnerships signal expansion beyond pure document processing. ABB Robotics invested in LandingAI's Series B in September 2025 to integrate LandingLens into robotics software, targeting 80% reduction in vision AI deployment time. LandingAI was also named Snowflake Startup Program Data Cloud Product Partner of the Year in June 2025, and in January 2026 launched ADE as a Snowflake Native App for energy sector applications.

Developer adoption is the near-term growth lever. A Financial AI Hackathon drew over 1,000 developers globally, producing winning solutions in loan underwriting, fraud detection, compliance automation, and invoice processing. A free DeepLearning.AI course taught by LandingAI staff covers ADE's visual parsing approach and an AWS production pipeline. ADE triggers on S3 uploads, loads parsed documents into Amazon Bedrock Knowledge Base, and queries them via Strands Agents, positioning ADE within the enterprise cloud stack without requiring a direct enterprise sales relationship. Adoption figures cited by the company (90% reduction in information search times, billions of pages processed) are vendor-stated and unconfirmed by independent sources.

How LandingAI processes documents

LandingAI's Agentic Document Extraction replaces the conventional OCR-then-LLM pipeline with a single visual AI pass. As the DeepLearning.AI course description states: "ADE treats documents as visual objects. It uses custom models to parse complex elements and ground extracted fields to precise locations on the page." The DPT-2 model reads document layout as a visual object while preserving spatial relationships between text, tables, and figures, keeping structure intact rather than converting the page to plain text first and losing it in the process.

Cem Dilmegani, author of the AIMultiple benchmark, noted that LandingAI "left traditional approaches behind and used OCR in different areas. Their document processing is not limited to one type of data extraction. They claim that their agentic document extraction tool can extract complicated images and 'fill in the blanks' when needed." The benchmark specifically confirmed that ADE "can extract complicated and mixed data (text and table on the same page) without any prompting," which distinguishes it from prompt-engineered general models.

The extraction pipeline produces structured Markdown and JSON with visual grounding: each extracted field is traceable to its bounding box in the source document. This cell-level provenance supports RAG pipelines and audit workflows. ADE Split handles multi-document PDFs by separating them using layout-aware visual AI before extraction begins.

The landingai-ade Python library on GitHub auto-splits and parallel-processes PDFs of 100 or more pages in a single API call, and has been tested on documents exceeding 1,000 pages. Integration requires 3 lines of SDK code across Python, TypeScript, and JavaScript. The platform is available via cloud, on-premises, and virtual private environment deployment, with a Zero Data Retention option. Pricing is credit-based per page processed; full details are at docs.landing.ai/ade/ade-pricing. Security certifications are referenced at landing.ai/security-at-landingai but specific certification names are not disclosed.

The unresolved confidence scoring question sits at the center of the production readiness debate. If the feature is live as the product page states, ADE closes a critical enterprise gap. If it is not, the platform remains a strong development and prototyping tool that requires additional validation layers before production deployment in regulated environments. Unstract, another open-source LLM platform for intelligent document processing, addresses this gap through explicit hallucination mitigation and token-level audit trails. Cambrion, a Munich-based agentic AI platform also founded in 2024, takes a comparable vision-language model (VLM) approach to zero-shot document processing without OCR, offering another reference point for teams evaluating this architectural pattern.

Use cases

Financial document processing

Financial institutions use ADE to process 10-K forms, financial statements, and regulatory filings. The platform's smart 10-K auditor implementation demonstrates visual grounding for audit trails and element traceability across complex nested tables. Hackathon winners extended this to loan underwriting, fraud detection, and invoice processing across multiple formats, currencies, and languages, combining the ADE SDK with AWS Bedrock in multi-agent architectures with RAG and deterministic rule engines.

Andrew Ng has framed the opportunity directly: "In financial services and in many other places, we have so much data. We retain the invoices, the financial documents, the 10Ks to the K1s, but we have so much data that for a long time has just been sitting around in our data warehouses or sometimes even on our laptops unprocessed."

Acuity Knowledge Partners represents the institutional demand side of this problem: the firm serves 800+ financial institutions with AI-powered document processing and research automation, illustrating the scale of unprocessed financial document backlogs that platforms like ADE target.

Healthcare document intelligence

Dr. Declan Kelly from Eolas Medical reported: "ADE has significantly outperformed other document extractors we've used. It has helped us build an Agentic RAG answer engine, based on unique healthcare institutional content, to offer instant, validated support to medical professionals at the point of care." The platform's visual grounding and source traceability are particularly relevant for clinical documentation where provenance matters for compliance.

Energy sector operations

The January 2026 Snowflake Native App partnership targets energy organizations managing large volumes of unstructured operational and regulatory documents. Deploying ADE as a Snowflake Native App brings document extraction directly into the analytics stack, eliminating the data movement that previously kept document intelligence separate from operational data. CEO Dan Maloney noted that critical operational and regulatory intelligence remains "locked outside the analytics stack" for most energy operators.

Manufacturing visual inspection

Through the ABB Robotics partnership, LandingLens is integrated into robotics software for quality control applications. Sami Atiya, President of ABB Robotics, stated: "Installation and deployment time is done in hours instead of weeks, allowing more businesses to automate smarter, faster and more efficiently." This use case extends LandingAI's footprint beyond document processing into physical production environments.

Fast.io positions LandingAI as best suited for "agents processing research papers, technical documentation, or documents that need contextual reasoning beyond simple extraction," while noting it is a "newer product with smaller user base" compared to established competitors like Amazon Textract and Google Document AI. LandingAI also serves finance, insurance, legal, and logistics industries with applications in risk assessment, regulatory reporting, and contract review.

Technical specifications

Resources

• LandingAI website
• Agentic Document Extraction
• LandingLens
• VisionAgent
• agentic-doc Python library on GitHub
• Document AI course from DeepLearning.AI
• AIMultiple agentic document extraction benchmark
• Deep Analysis: LandingAI and the future of IDP
• LandingAI competitive analysis
• Agentic document processing guide
• LangExtract: Google's open-source Python library for structured extraction from unstructured text using LLMs, offering a complementary approach to ADE for developer pipelines
• Agentic capability overview

Company information