LandingAI

Confidence 0.75 · 1 source · last confirmed 2026-06-09

LandingAI is the computer-vision / document-AI company Andrew Ng founded (described in the webinar as “about seven years ago”; ~2018). It is the wiki’s first document-intelligence vendor entity, promoted on 9 June 2026 from a Dangling mention (long referenced on the Andrew Ng entity as his company) to its own page when the wiki ingested its first LandingAI-channel source — the [[2026-05-26-landingai-touchpoint-to-outcome-front-office-processes|Touchpoint to Outcome webinar (26 May 2026)]].

The company self-describes as “vision-first AI” — distinct from the “OCR + LLM” approach it positions against — and notes a history of “hundreds of projects to many enterprises” on visual / unstructured data before pivoting to its current flagship product about a year before the webinar.

Role in the wiki

LandingAI anchors the document-intelligence vendor altitude — the technical-and-commercial vantage on extracting high-accuracy, grounded structured data from unstructured enterprise documents. Where Google’s Gemini File Search (multimodal RAG) approaches verifiable document retrieval from a RAG / hyperscaler-API angle, LandingAI approaches it from a vision-first structured-extraction angle. It is also the commercial instantiation of Andrew Ng’s “unbig in AI” / fit-for-purpose-beats-generality thesis at the document-AI altitude.

Agentic Document Extraction (ADE)

LandingAI’s flagship product. Stated differentiators from the webinar:

• Vision-first. Reads a page as an image — identifying blocks, structures, and human reading-order — then extracts in that sequence, rather than treating the page as pixels + text. Pitched as the reason competitors “stuck in pilot form” on complex documents (nested tables, handwriting, signatures, graphs, mixed formats) succeed with ADE.
• Visually grounded. Every extracted value references back to the source cell / word / figure / page, giving a programmatic + UI audit trail that suppresses hallucination — “financial services and life sciences love that type of technology… auditability of data is paramount.”
• Zero-shot. “No training needed, no fine-tuning needed” — point documents at the API and receive structured output. API primitives include parse, extract, classify, plus schema-driven selective extraction and document splitting.
• Built for developers as part of an enterprise-grade agentic stack.

DPT — Document Pre-trained Transformer

LandingAI’s proprietary model family underpinning ADE. DPT2 is shipping; DPT3 was described as “just about to be launched” (as of 26 May 2026). The company cites an “independent benchmark” (not shown in the webinar) placing ADE at “the top of the 99s plus accuracy… even above human performance” — a vendor-reported, uncorroborated claim. (Currently a Dangling concept/product mention; promote DPT to its own page on a second source.)

Deployment options

• ADE Cloud — multi-tenant SaaS, hosted in the US or EU, optional zero-data-retention.
• ADE VPC — virtual private cloud, deployed into the customer’s own AWS / Azure hyperscaler tenancy.
• On-prem / air-gapped — for government and high-control financial organisations requiring data in their own data centre. More prerequisites; available on request.

Customer evidence (vendor-reported)

• A global bank adopted ADE for KYC / client due-diligence / fraud detection across diverse unstructured documents, moving from “hundreds of analysts… thousands of hours” of manual review toward production automation; LandingAI claims it is “adding two or three [banks]” to its portfolio monthly.
• Applicable verticals named: insurance, pharmaceutical / life sciences, engineering, manufacturing.
• Trust posture: trust.landing.ai; customers across North America, Europe, and Asia.

Partner ecosystem

• TCG / OCTO (Dangling) — TCG (est. 1996) embeds ADE inside its OCTO no-code process-automation platform, orchestrating extraction with the “before and after” (input normalisation, validation, grounding, downstream-system connection) to reach “100% accuracy” before systems of record are touched. The webinar’s insurance / car-rental demos are OCTO-driven. See the source page.

Open questions

• DPT benchmark provenance. The “independent benchmark” and “above human performance” claims are vendor-reported and unshown; corroboration (the cited blog, a third-party eval) would lift document-intelligence confidence above the vendor cap.
• Founding date. Webinar says “about seven years ago” (→ ~2019) but LandingAI is generally dated to 2017–2018; pin against a primary source.
• ADE vs LandingLens. LandingAI’s older industrial-vision product (LandingLens) is not mentioned in this webinar; the relationship between the visual-inspection lineage and the document-AI product line is untracked.
• TCG / OCTO promotion. Promote TCG, OCTO, and Neil Walker on a second source.

Mentioned in

• 2026-05-26-landingai-touchpoint-to-outcome-front-office-processes — the founding source: LandingAI’s own ADE webinar with partner TCG/OCTO.
• Andrew Ng — founder entity; LandingAI listed in his affiliations and the “unbig in AI” / industrial-vision lineage.