Skip to main content
appraisal-OS home
Meet Aria

The copilot that cites her work.

Aria sits inside every field — reading the contract, pulling comps, drafting narratives, checking UAD 3.6 enums. Every answer comes with a clickable citation chip. If she can't be sure, she asks. She doesn't guess values that carry reviewer risk.

100%
Cited claims
0
Blind AI guesses
RAG
Grounded retrieval model
You
Approve every value
The workflow

How it works, step by step.

  1. 01
    Reads the engagement letter

    Parses loan type, intended use, effective date, and special instructions. 1004 header pre-fills.

  2. 02
    Pulls source documents

    Assessor card, MLS sheet, flood cert, contract, prior appraisal — indexed and searchable.

  3. 03
    Proposes values with citations

    Every field she touches shows the source you can click. Nothing is eyeballed.

  4. 04
    Drafts narratives with sources

    Neighborhood, site, and comparable commentary — each claim tied to a citation chip.

  5. 05
    Checks math + compliance

    ANSI Z765 GLA, HPI time adjustments, MISMO 3.6 enums, conditional unlocks.

  6. 06
    Asks when she's unsure

    Low-confidence fields surface as a question for you, not a silent guess.

What's included

Everything in this feature.

Grounded retrieval (RAG)

Every answer is pulled from the documents and data sources attached to the order — not from open training data.

Citation chips

Click any Aria-proposed value to see the source document, page, or data point.

UAD 3.6 enum coach

For every UAD 3.6 field, Aria explains the definition, the allowed enums, and proposes the right value.

Narrative drafts

Neighborhood, site, and comparable paragraphs with in-line citations.

QC assistant

Flags inconsistencies across the URAR (GLA vs sketch, sale price vs engagement letter, etc.).

Graceful refusal

When she can't be sure, she asks. Silent hallucinations are a reviewer nightmare — she doesn't do them.

Why this matters

AI that cites its work changes what reviewers see.

A reviewer opening a report grounded in clickable citations spends less time suspicious and more time confirming. That's the entire point — reduce friction on the way out the door, not add more fireworks to the way in.

FAQ

Quick answers.

  • Aria uses Anthropic's Claude family for reasoning, retrieval-augmented generation (RAG) over your order documents, and a handful of task-specific tools for math and compliance checks.

Ready to see it live?

Free for the founding cohort. No credit card. You can be working on a real order in under two minutes.