A desktop workspace for UAD 3.6 appraisal reports.
A Windows and Mac workspace for writing UAD 3.6 appraisal reports, the redesigned data standard from Fannie Mae and Freddie Mac. A Jarod Vyent LLC product, with Professional Appraisal Services as the pilot firm.
proappraisals.us: more than five new clients within 30 days of launch.
Report editor
- Runs locally in the app
- Report access validated online
Authorize and save
- Online report access validation
- Cloud operations go through the API
Authorize
- Account, firm, license and seat
- Checked on every request
AI first pass
- Drafts empty sections only
- Metered against the firm
Administer
- People, roles and seats
- Invitations, from a browser
Identity Platform
- Verified email sign-in
- Account rechecked on each call
Cloud SQL
- Private network, no public IP
- Revisions cannot be changed or deleted
Model Armor
- Screens every drafting call
- Refuses the draft if unreachable
Every report operation is authorized on the server, on every request.
- The appraiser uses the desktop editor with online access validation; prepared fieldwork notebooks support offline collection.
- Saving sends the report to the API, which checks the account, firm membership, license and seat before it reads or writes anything.
- A retained report snapshot and PDF are stored as a revision that the application cannot change or delete.
A retained report revision has to stay as it was recorded.
Appraisal work needs a reliable record. Once a report snapshot and PDF are retained, that version has to survive unchanged, and only people the firm has given a seat should be able to open it.
The server and the database enforce both rules. Every report operation checks for a verified account, membership in the firm, a valid license and an assigned seat, on every request. A database trigger refuses any update or delete of a stored revision.
A desktop editor, with access controlled by the cloud API.
The report editor runs inside the desktop app on Windows and Mac. Report access requires online account, license and seat validation; AI and cloud saving also need a connection. Prepared fieldwork notebooks support bounded offline collection before their evidence is reviewed and imported into a report.
The API runs on Cloud Run over a private Cloud SQL database, provisioned in Terraform with a test that fails if the database is ever given a public address. Firm owners manage people, roles, seats and invitations from a browser, and authoring stays in the desktop app.
AI drafting that never overwrites the appraiser’s text.
The AI first pass drafts narrative sections from what the report already holds. It only writes into sections that are empty, it skips anything the appraiser wrote, and it says which sections it wrote so the appraiser knows what to check before signing.
When the report holds nothing a section could be written from, the model leaves it blank and says why, instead of writing a paragraph about the missing data. Google Cloud Model Armor screens every drafting call on the way in and on the way out, and if the screen is unreachable the draft is refused. Only owners, administrators and appraisers can start a pass, because every pass is metered against the firm.
In pilot with one firm.
The cloud side and AI drafting are deployed on Google Cloud, and Professional Appraisal Services is the pilot firm. Desktop builds and the iPhone/iPad fieldwork app are distributed through controlled pilot and testing workflows. Licensing is built in; commercial billing and broader availability have separate rollout gates.
Outcome figures for the app wait until the pilot has put real reports through it. The documentation center covers the applications, AI models, architecture and current release boundaries.
A new site for a 25-county appraisal practice.
I also replaced the firm’s old website. The new proappraisals.us has five service pages, a map of the 25 Ohio counties the firm covers, a request form that sends each request to the office, and a product page for appraisers. It is a Next.js app on Cloudflare Workers, and it went live on 11 September 2026.
How the numbers above were counted. The traffic figures cover the first 30 days, 11 September to 10 October. Cloudflare counted 6.0k unique IP addresses, and most of them were crawlers and vulnerability scanners. Counting only addresses that behaved like a person’s browser, by loading the site’s scripts and images, leaves about 300 visitors: 267 distinct addresses from 16 September on, plus up to 63 from launch week that can no longer be matched against later days. That filter is conservative: a person on a mobile network address that a crawler also used is left out, so the count is a floor. Cloudflare’s own dashboard shows a rolling 30 days, so its totals move from day to day; these are fixed to the first 30. The client count comes from the firm. The Lighthouse figures come from Lighthouse 12.8 runs against the home page on 15 October. On a phone profile, performance scores 94 to 95 and SEO and best practices stay at 100. Accessibility scores 91 on both, held back by one low-contrast element and some small touch targets.