139 N Ocean Ave, Suite 204, Patchogue, NY 11772Mon to Fri 9 AM to 7 PM, Sat 10 AM to 2 PMHablo español

(631) 823-7630

Software · built in-house at Yanza Brokerage Group

Yanza Desk

An AI-assisted sales desk and CRM for an independent New York insurance broker. It turns public records into a ranked call list, checks every call, letter and email against the rules before it goes, and keeps client records on the broker's own computer.

I built it, and I run my brokerage on it every day. It isn't for sale yet. Every screen on this page shows made-up sample data.

Sample data The Desk's Today screen: the next call with three dated reasons, goals for the day, meetings and an inbox, all with sample data
Today: one next step, and the reasons why.
Automated tests
3,600+run before every change
Code and tests
250,000+lines, Python and JavaScript
Public-record sources
15federal, state and city
Required packages
0Python standard library only

Why I built it

One broker, a lot of rules, no back office

I opened Yanza Brokerage Group in Patchogue in October 2026. An independent broker has to find businesses to call, call them inside the Do Not Call and calling-hour rules, write to them without a single misleading word, and keep their records private under New York's cybersecurity regulation, 23 NYCRR Part 500. Large agencies have staff and compliance teams for that. I have one desk.

So I wrote the rules into software. The Desk decides who to call next and why, refuses a call or an email that breaks a rule, and explains the refusal in plain words. Nothing it shows a prospect is made up: it uses public filings and real quotes only, and its own estimates stay private, labeled as estimates.

How it works

Public records in. Rule-checked outreach out.

Five parts, one database, one page in the browser.

1. Lead lists from public records

The engine reads 15 public-record sources, such as U.S. Department of Labor Form 5500 benefit-plan filings, NY Department of State corporation filings, FMCSA trucking records, OSHA inspections, SBA loans, liquor licenses and New York City permits, and matches them onto one record per company.

Each fact is labeled actual, projected or estimate, with its source. Ranking puts the companies most worth a call today at the top, each with a dated reason.

Sample data Contacts: saved lists and filters for stage, product gap, county and size, with sample companies
Contacts: saved lists and filters.

2. Calling, checked before every dial

Call Mode shows one company at a time with a script and one-key results. Before a number can be called, the Desk runs five checks, and the first one that fails decides:

  1. Opted out, or on the do-not-contact list?
  2. A valid number?
  3. Do Not Call list for that area code loaded, and fresh?
  4. Inside calling hours where they are?
  5. Fewer than 3 tries in 24 hours?

The Desk never dials by itself. A call starts only when I click it, through my business phone app. Automated tests read the code and fail if anyone, the AI included, adds auto-dialing, a recorded or AI voice, or text sending.

Sample data Call Mode refusing a sample number because its area code's Do Not Call list is out of date
The call check refuses a number whose Do Not Call list is out of date.

3. Outreach that passes a wording check first

Letters, one-pagers and emails go through a wording checker built from New York insurance advertising rules. A price superlative, a savings promise or a promise of results is a hard stop with no override, and the Desk points to the rule.

Autopilot can send a short, checked email sequence, and it starts switched off. Before each send it re-checks every gate: the opt-out register, the do-not-contact and bounce lists, the wording check and footer, the sending window, daily caps and anything held for my review. One Pause button stops it all.

Sample data The wording checker blocking a price superlative in a sample one-pager, with the rule it breaks
The wording checker blocks a price superlative, with no override.

4. A CRM and deal desk built for a brokerage

Every company has one page: people, numbers, filings, calls, quotes and documents. A Client view hides private numbers when someone else can see the screen, and it's enforced on the server, not just hidden on the page.

The deal desk follows a reply to a quote, a win and enrollment. When a deal is won, the Desk prepares the follow-up paperwork; I confirm the dates and press Send myself. Automations make the next task or draft and never send, text or dial on their own.

Sample data A sample company page in Client view, with contacts, stage, activity counts and the next step
A company page in Client view: private numbers hidden.

5. Employee benefits from Form 5500 filings

Employers that sponsor a benefit plan file a Form 5500 with the U.S. Department of Labor every year, and the filings are public. The benefits module reads them for my New York territory and ranks employers by plan renewal date, plan size and what coverage the filings show.

Anything an employer sees uses only the facts in their own filing, with the plan year. The Desk's cost estimates are for my planning and never appear in outreach.

Sample data Settings, Automations: built-in rules that create the next step and never send, text or dial, with sample results
Automations make the next step and never send on their own.

Demo

See it in about a minute

A recorded walkthrough of the Desk with a made-up sample agency. No real person, business or client appears.

Yanza Desk demo, sample data, 1 minute 10 seconds, with captions in the video.

AI assistance

An AI assistant that asks first

I work with Claude, Anthropic's AI model, beside the Desk. It reads my day and tells me who to call and why, explains a Desk problem in plain words, drafts letters from my templates and walks me through setup. It logs a call or a note only after I say yes to exactly what it will record.

It reads the Desk through the same rules and privacy filter as the screen. A guard blocks it from opening the database, the client folders or any key, and it never sends, dials or texts.

How it's built

Small stack, lots of tests

The Desk is a local web app: a Python server using only the standard library, SQLite, and plain JavaScript with no framework or build step. More than 3,600 automated tests run before every change. I designed every rule and reviewed every change; Claude wrote much of the code with me in Claude Code.

  • Python 3, standard library
  • SQLite
  • Plain JavaScript
  • Claude (Anthropic)
  • Windows DPAPI and macOS Keychain
  • Public datasets: DOL, NY DOS, FMCSA, OSHA, SBA, NYC Open Data

The data sources are national, so the same Desk can be set up for another state. A version other independent agents can run on their own computers, each with their own leads and records, is the next step.

Questions about the Desk

Ask me anything else at contact@yanzabrokeragegroup.com.

Can I buy or download the Desk?

Not yet. Today it runs one agency, mine. If you're an independent agent and want to hear when that changes, or you'd like a live walkthrough with sample data, email me.

Is my information in the Desk if I'm a client?

Your records are stored on my office computer, not in a cloud database, and handled under my privacy policy. Nothing on this page or in the demo shows a real client. See the privacy policy.

Does the AI make calls or send messages?

No. The assistant reads, explains and drafts. Calls start only when I click them, and the Desk never sends a text. The one exception to "I send it myself" is Autopilot email, which is off until I turn it on and re-checks every rule before each email.

Where does the lead data come from?

From public government records, such as Form 5500 filings, state corporation filings, federal trucking and safety records and New York City permits, plus quote requests people send me and leads I buy from lead services. Every fact keeps its source.

Want a walkthrough?

I'll show you the Desk with sample data, on a call or in person in Patchogue.