Michael Ratnikov
All case studies

Quote Bot via Rendered SourceInsurTech

Insurance CRM that passed SOC 2 and HIPAA audits

A CRM and sales platform for US life-insurance agencies. It holds personal and health data. I was lead architect in a team of five for five years, from the first release through the audits.

Passed SOC 2 and HIPAA audits5 years, no rewritePer-agency data isolation5 AI features
Insurance CRM on laptop and phone
The challenge

The client was replacing its old system. Insurance agencies would keep their customers' personal and health data in the new one, and the platform had to pass SOC 2 and HIPAA audits. Security had to be part of the first release, not added in a rush before the audit.

I worked on it through Rendered Source, a US software company.

I have been through SOC 2 and HIPAA audits on a live platform. In a Review, I check your product against the same controls those auditors checked.

Data it had to protect
  • Social Security and tax numbers

    Kept on customer and policy records

  • Health answers

    From the insurance interview

  • Signed applications

    Back from e-signature, stored per agency

  • Call recordings

    From the phone built into the CRM

HIPAA for the health data

SOC 2 for all of it

What the auditors checked

The same list I check in your product

Auditors asked how the platform protected data. Each answer below was built into the product from the first release. Under each one: what I look for in your product.

Customer data kept apart

Each agency had its own database, logins, app and web address. No table held two agencies' customer data.

In your product I check

Whether one customer can see another's data. In AI-built apps, one missing line is often enough.

One agency, one environment

Agency A

  • Database
  • Logins
  • App
  • Web address

Agency B

  • Database
  • Logins
  • App
  • Web address

No table holds two agencies' customer data

Admin console

Create · Release · Suspend · Delete

Who can see and change what

Three roles, with rules for own, team and team-lead records, enforced inside the database queries. No screen can show a row the user may not see.

In your product I check

Whether access rules live on the server, or only in the screens.

Secrets and API keys

26 secrets in AWS Parameter Store, none in the code. A key can change without a new release.

In your product I check

Keys and passwords in the code, in its history, or in the browser.

A record of every change

Every change logged: who, when, before and after. Opening health data needs a reason, and is logged too.

In your product I check

Whether you can answer "who changed this, and when?"

Known security flaws

The release pipeline blocks a release with a known critical or high flaw that has a fix.

In your product I check

Old libraries with known flaws, and what stops them reaching production.

Backups and rollback

Encrypted daily backups. A new version that does not start is rolled back automatically.

In your product I check

Whether a backup was ever restored, and how a bad release is undone.

Every release can be undone

New release

One version, one agency at a time

Gate 1: database changes first

If they fail, the release stops here.

Gate 2: the new version must start

It has to start and respond.

Automatic rollback
if it does not start

Live for this agency

Then the next agency

Agency A: updatedAgency B: updatingAgency N: waiting

AI that stays inside its limits

The chat bot can only pick an answer the agency wrote. No match gives a safe fallback, not a guess.

In your product I check

What your AI features can say or do on their own.

The AI picks the answer. People wrote it.
Visitor asks a question

ChatGPT

Picks an answer, or asks a clarifying question

Every message passes all three

  • Approved answers only

    Reply text comes from the agency's own list

  • Strict reply format

    Checked by code, creativity set to zero

  • Full chat history

    Loaded from the CRM every time

Agency-approved answer sent

No match? A safe fallback message, never a guess.

What the system costs to run

I moved the deal-scoring model off a paid, always-on AWS endpoint into the app. That bill is gone.

In your product I check

Monthly cloud and AI services you do not need.

How does your product do on this list?

Book a 30-minute Risk Assessment Call. You get your 3 biggest risks in writing after the call.

How the work ran

Five years of change on one foundation

5 years

on one platform

659

database changes

.NET 8

upgraded from .NET Core in place

0

rewrites

Options, not orders

Goals came from the CEO. I came back with options, trade-offs and a budget view, then built the one the CEO chose.

Small, safe releases

Releases went out one agency at a time. A failed database change stopped the release, and a version that did not start was rolled back.

Rules in writing

I wrote the architecture, security, testing and review standards into the repository. The team and the AI coding tools read the same rules.

This is how I fix a product after a Review. Your stack stays. Fixes go out in small releases you approve, and every release can be rolled back.

The product

What agents, customers and agency owners got

For agents

  • Calls, SMS, video, email and live chat on one timeline per customer
  • A shared team inbox, updated live
  • Two-way email sync with Gmail, Outlook and other mailboxes
  • Sales pipelines with a win probability on every open dealAI
  • Policy cases with an underwriting checklist and documents
  • Call transcripts with summariesAI
  • Meeting summariesAI
  • An assistant that helps agents fill in formsAI

For customers

  • Buy a life-insurance policy online, without an agent
  • Sign the application online, insured and payor each with their own link
  • Chat on the agency's website, answered only with approved answersAI
  • Card numbers never touch the CRM: the payment token goes straight to the insurer

For agency owners

  • Their own environment, created and released from one admin console
  • Dashboards agents build themselves, with CSV and Excel export
  • Chat bot answers the agency writes, tested before go-live
  • Email campaigns from a template library
  • A log of every change, every login and every look at health data

Technologies

Application

.NET 8ASP.NET CoreAngularPostgreSQLSignalR

Cloud

AWSECS FargateAurora ServerlessCognitoCloudFrontKMSWAFSQSCloudWatchTerraformGitLab CIDocker

AI

ChatGPT APIML.NETAmazon TranscribeAmazon TextractAmazon Lex

Integrations

TwilioAmazon ConnectAmazon ChimeMailgunGmail APIOutlookDocuSignStripe
Review

What the client said

“Michael is in the top 1% of any architect we have worked with.”

“Working with Michael is in the top 1% of any experience my company has had with hundreds of developers. He is a leader with good judgement, highly technical, very kind and as honest as they come. I don't think I could recommend anyone more highly. The technical abilities were of the highest caliber, but architecturally and feature suggestions saved a lot of money and time. I have a software development background and the end results we achieve are in the top echelon of any system I've been involved with.”
Paul Janisko

Paul Janisko

CEO, Rendered Source

Is a deadlinecoming?

A round, an audit, a big customer, a vendor contract. Book a 30‑minute Risk Assessment Call. We go through who built your product, what must be ready before that date, and what breaks today.

Free. No preparation. You leave with your 3 biggest risks in writing, and an honest answer on whether a review is worth it now.