Technology
Areal isn't a generic document tool retrofitted for lending. It's a purpose-built AI platform — trained on millions of real mortgage and title pages, running on modern AWS infrastructure, API-first from day one.
99%
Accuracy on critical fields
1,500+
Document types supported
4,000+
Data points per loan
2 sec
Typical API response
Patent Pending
Areal CD Balancer and Areal Copilot Agent are patent pending — the balancing engine and the agent execution bench that runs on top of it.
Read the announcement →Trusted by Top-Tier Mortgage & Title Companies
Our Core Differentiator
Areal isn't a wrapper around a general-purpose LLM. Three engineered layers — proven over years and millions of mortgage and title pages — work together so the platform behaves predictably on the documents that actually matter.
Layer 3
Scoped, traceable agents execute mortgage workflows — every step constrained, planned, and logged before it runs.
Layer 2
Reads from and writes back to Encompass, MeridianLink, Byte, and TPS — data stays in sync with the system of record, not siloed.
Layer 1 · The Foundation
Transformer models trained on millions of labeled mortgage and title pages — 1,500+ document types, 4,000+ data points per loan.
To deliver a platform that is
99% accuracy on critical fields, proven in production at scale.
Every extraction and agent action is logged, traceable, and reviewable.
The same rules run on every loan, every closer, every month-end peak.
No hallucinations. No false judgments. Scoped inputs, constrained execution, and full traceability keep the AI inside guardrails — extracting and verifying what's on the page, never guessing.
The Stack
Every layer of the platform is designed around the realities of mortgage and title workflows — not adapted from a general-purpose document tool.
Purpose-built on modern AWS architecture for scale, reliability, and security — processing millions of pages weekly without downtime.
Document understanding runs on transformer architectures and large language models, fine-tuned specifically for mortgage and title language and structure.
Every capability is exposed via API. Documents are classified and extracted in near real-time — typically within seconds of upload.
Direct, native connections to ICE Encompass, MeridianLink, Byte, RamQuest, ResWare, and SoftPro — no middleware layer required.
How Documents Move Through Areal
1
Documents received from LOS, TPS, or direct upload
2
Every page auto-classified across 1,500+ document types
3
4,000+ data points per loan pulled at 99% accuracy on critical fields
4
Agents act on extracted data — balancing, review, verification
5
Results write back to your LOS or TPS automatically
Our Data Moat
Model quality in mortgage and title isn't just an algorithm problem — it's a data problem. Areal has spent years building the labeled dataset that makes 99% accuracy possible.
1
Every page adds to a proprietary labeled dataset spanning the full range of mortgage and title document types.
2
Models are trained specifically on mortgage and title language, formats, and edge cases — not adapted from generic OCR.
3
Areal Copilot Agent, launched October 2025, was the industry's first agentic AI platform built specifically for mortgage.
Millions
of pages processed every week, compounding the data advantage
2020
Launched
99%
Accuracy today
Trust & Security
Independently audited security controls
Every extraction and action logged for compliance
Data encrypted in transit and at rest on AWS
Built with regulatory and investor requirements in mind
Core capabilities
CD Balancer and Copilot Agent — both patent pending — are built on the same document engine. These are the pieces that do the work.
Every page identified as it lands — 1,500+ document types, duplicates caught, bundled PDFs split automatically.
Learn more →
4,000+ data points per loan, at 99% accuracy on the critical fields — including signatures, stamps and notary blocks.
Learn more →
Every value marked on the page it came from, so review is a glance and the audit trail is built in.
Learn more →
The reasoning layer — values compared across documents, the LOS and third-party systems, returning exceptions instead of data.
Learn more →
Get Started
We'll show you exactly how the extraction and automation engine performs on your own loan files.
