Practice-area sites, intake, and a guardrailed AI concierge for a plaintiff-side law firm, deployed into the firm’s own AWS account. iSimplifyMe-built. iSimplifyMe-operated.
What is the regulated law-firm fleet?
iSimplifyMe builds and operates a fleet of practice-area websites for a plaintiff-side law firm inside the firm’s own AWS account: one Next.js and SST codebase forked per property, production deploys dispatched by a person after the firm’s review, contact intake through the Apex leads endpoint with spam classification in enforce mode, and an AI concierge under the regulated-tenant guardrails. The firm is not named; references are available under NDA.
The setup
A plaintiff-side law firm wanted its practice areas on separate properties, each with its own intake, its own concierge, and its own paid-media landing pages — without handing its client data to a vendor’s cloud account and without a marketing team maintaining code. Every AI surface had to meet the regulated-tenant floor iSimplifyMe applies to legal and medical concierges.
The mandate: build the fleet inside the firm’s own AWS account, launch each property only after the firm has reviewed it, and operate the properties afterwards.
Architecture
One codebase, forked per property. Each site is a Next.js application deployed with SST to Lambda behind CloudFront, CloudFront, S3, and DynamoDB in the firm’s account. Guards, tests, and fixes are ported across the forks, so the fleet moves together. A merge is not a deploy: production is a workflow_dispatch a person triggers after the firm’s review; staging deploys on every merge.
Launch gate. Robots stay closed until sign-off. A property’s sitemap and its no-index decisions read the same rule, so a page cannot be submitted and excluded at once. Gate tests are inverted at launch: they now fail on an accidental un-launch the way they failed on an accidental launch.
Intake. Forms post to the Apex leads endpoint (/api/leads/submit): one call stores the lead, classifies it, and sends the firm’s notification from iSimplifyMe’s verified mail domain, so no property needs its own mail setup. The classifier runs in enforce mode — a spam submission receives a normal response, is stored flagged, and sends nothing.
AI concierge. A per-tenant persona in S3 with a system-guardrails block the model reads on every turn, a disclaimer band the widget shows on open, and hotline tokens. The assistant answers what the site covers, never evaluates a case, never names a filing window, and routes anything case-specific to the firm’s intake. Observability is per tenant through Apex: uptime, response time, bot traffic, and the leads pipeline itself.
Why this is harder than it looks
A site’s scope is not the firm’s licensure
One property is deliberately scoped narrower than the firm’s licensure. The first concierge persona said the firm could not take matters outside that scope — accurate about the site, false about the firm, and it turned a real caller away. The fix separated the two statements: the assistant explains what this site covers and offers the callback for everything else. Scope rules on a regulated surface are reviewed against the client’s actual licensure, not the site’s content plan.
A fresh AWS account throttles a fleet
A new account ships with a Lambda concurrency quota of ten, shared by every function in it. Six production stacks and more than forty functions would have throttled each other under the first traffic burst. The quota was raised to one thousand and verified against the service-quotas API before launch. It is now a standing check for every client-owned account.
Deploys never delete what they stop shipping
Serverless deploys upload new static assets but never prune the old ones, so a removed image or page stays live at its old URL through the CDN indefinitely. The fleet treats every retirement as an explicit deletion, and verification checks the HTTP status of what the rendered page references, not whether the build passed.
Registrar records fight the first deploy
SST can overwrite an existing DNS record of the same type, but it cannot turn a registrar’s apex A record into the CNAME a CloudFront distribution needs. The first production launch stalled on exactly that. The runbook now deletes the registrar’s apex and www records before the first production deploy, then verifies the name servers and the edge headers independently.
What’s running in production today
| Component | State | Notes |
|---|---|---|
| Practice-area property 1 | Live | Production since August 2026; indexing opened after the firm’s sign-off; public routes indexable, self-canonical |
| AI concierge | Live | Regulated-tenant persona and disclaimer band; case-specific questions route to the firm; served through Apex |
| Contact intake | Live | Apex leads endpoint; spam classifier in enforce mode — the visitor gets a normal response, the firm never gets the spam |
| Paid-media landing pages | Built and deployed | Two pages live with server-side conversion events to the ad platform, lead capture on the firm’s own CRM; a third staged |
| Further practice-area properties | Staged | Built, deployed to staging, robots closed; each opens on the firm’s review |
All infrastructure in the firm’s own AWS account. Verified at the edge after each deploy — name servers, response headers, robots, sitemap, and the HTTP status of every asset the rendered pages reference. Sites, intake, and concierge iSimplifyMe-operated; landing pages built and deployed.
Build log
- Phase 0
Regulated-tenant guardrails become the floor (May 2026)
System-guardrails block and disclaimer band become the floor for every legal and medical concierge tenant iSimplifyMe serves. Every property in this fleet inherits it at launch.
- Phase 1
New property forked and staged (August 2026)
One codebase, forked per property, deploying through GitHub Actions into the firm’s AWS account with SST. Shared CI guards — fleet lint, an AI-surface content gate, mobile-geometry checks — ported across the fleet so a fix in one property reaches all of them.
- Phase 2
Landing pages with server-side conversion events (August 2026)
Two paid-media landing pages built and deployed with the ad platform’s conversions API wired server-side and verified with test events, so attribution survives browser privacy controls. Lead capture runs on the firm’s own CRM.
- Phase 3
Production launch and intake enforcement (August 2026)
Name servers moved, first production deploy dispatched, concierge tenant wired, robots opened, sitemap and indexability rules unified into one source. Verified independently at the edge, not from the deploy log. Intake spam classification then moved to enforce, proven end-to-end on production: a gibberish submission gets a normal response, is stored flagged, and sends nothing to the firm.
- Phase 4
Remaining properties (staged)
Built and deployed to staging behind the same gates. Each opens to the public on the firm’s review, one dispatch at a time.
Built and operated, not delivered.
Most law-firm site engagements end with a handoff: a theme, a plugin list, and a chatbot nobody has read the prompt of.
This fleet runs in the firm’s own account, behind the firm’s own review, with intake, concierge, and monitoring operated by the people who built it. The same pattern deploys into any regulated client’s AWS account on the same SST-wired stack.
Frequently asked questions
Why deploy into the client’s AWS account instead of iSimplifyMe’s?
Because the account is the compliance perimeter. Everything the sites collect — form submissions, chat transcripts, lead records — lands in infrastructure the firm owns, under the firm’s IAM and billing. iSimplifyMe operates inside that perimeter through CI credentials the firm can revoke. Nothing about the engagement depends on the firm trusting iSimplifyMe’s account.
How is an AI concierge kept inside the rules for a law firm?
Every regulated tenant carries a persona with a system-guardrails block the model reads on every turn and a disclaimer band the widget shows when the chat opens. The guardrails say plainly that the assistant is not a lawyer and not the named attorneys, that no attorney-client relationship is created, that it never confirms someone has a case, never names a filing deadline, and routes anything case-specific to the firm. The floor has been live across iSimplifyMe’s legal and medical tenants since May 2026.
What does “gated behind the firm’s review” mean in practice?
A property is built and deployed to staging, but robots stay closed and production deploys are triggered by a person, not by a merge. The firm reviews copy and positioning. When it signs off, one dispatch deploys production, the DNS moves, and robots open. The sitemap and the indexability rules come from one source, so a page cannot be listed and no-indexed at the same time. Unreviewed properties never reach the public internet.
Who operates the fleet after launch?
iSimplifyMe. Each property is an Apex tenant: uptime, response time, and bot traffic are observed per tenant, the leads endpoint stores and classifies every submission, and the same CI guards run on every change. The firm gets the notifications; it does not inherit a codebase to maintain.
Can the firm be named?
No. The engagement is documented here by its setup type, not its client. References are available under NDA in qualified engagements.
Discuss a regulated site fleet in your own account
If your firm or practice needs its properties, intake, and AI surfaces inside an AWS account you own — and reviewed before anything goes public — we can walk through the pattern for your structure.
- Discovery call30 min · Free · No deck — actual mechanics
- PerimeterYour AWS account, your IAM, your billing
- iSM-operatedDeploys, intake, concierge, monitoring — all on us