Serverless pipeline that turns a Stripe payment into a live, verifiable license page for a life insurance agent — with no human in the loop.
An agent signs up, pays, and minutes later has a site at https://{name}.verifyanagent.com showing every state they're licensed in, their certificates, and one-click links to each state's official license lookup. Consumers use it to check that an agent is real before buying a policy.
- Live demo: https://judewakim.verifyanagent.com
- Get one built for you ($300, one-time): https://verifyanagent.com/get-started
Don't want to run AWS yourself? That's what the paid service is for. This repo is the engine; the hosted version is set up, reviewed, and maintained for you.
Text version of the diagram
Signup form ──POST /signup──▶ Admin/Signup Lambda ──▶ Airtable (record: submitted)
│ │
│ presigned PUT (images, PDFs) ▼
└──────────────────────────▶ S3 (private uploads)
│
▼
Stripe Checkout (client_reference_id = Airtable record ID)
│ checkout.session.completed
▼
API Gateway ──▶ Build Lambda
1. Verify Stripe signature
2. Load record by ID (not email)
3. OCR each license cert (Claude Haiku vision, structured output)
4. Build per-state license cards (50-state lookup map)
5. Render Jinja2 template ──▶ S3 ──▶ {slug}.verifyanagent.com
6. Regenerate sitemap.xml + public agent directory
7. Slack alert if anything needs human review
| Component | Files |
|---|---|
| Stripe webhook entry point | handler.py, stripe_verifier.py |
| Shared build pipeline | builder.py, site_generator.py, templates/ |
| Certificate OCR + name check | cert_ocr.py |
| 50-state license database | state_data.py |
| Asset handling | s3_manager.py |
| Admin portal + public signup API | admin_api.py, admin/ |
| SEO | sitemap.py |
| Operator alerts | notify.py |
Stack: Python 3 · AWS Lambda · API Gateway · S3 · SSM Parameter Store (KMS) · Stripe · Airtable · Anthropic API · Slack webhooks · PowerShell deploy scripts
- Webhook authenticity. Every Stripe event is signature-verified before any work happens; unsigned or forged requests are rejected immediately.
- Payment joined by record ID, not email. Stripe's
client_reference_idcarries the Airtable record ID, so a customer paying with a different email can't break or hijack a build. - Admin auth. Single operator account. The email comes from an env var and fails closed if unset. Passwords are PBKDF2-SHA256 (200k iterations, per-password salt) stored as a KMS-encrypted SecureString in SSM, never in S3 or code. Sessions are stateless HMAC-signed bearer tokens that expire after 12 hours, compared in constant time.
- Least-privilege public routes. Unauthenticated signup routes can only touch records still in the
submittedstate; once a record is paid, only the authenticated admin API can change it. - Direct-to-S3 uploads. Images and certificate PDFs go up via short-lived presigned URLs, so file bytes never pass through the API.
- No secrets in the repo. All credentials live in Lambda environment variables (see
.env.deploy.example). - Spam guard. Honeypot field plus optional Cloudflare Turnstile verification (
docs/turnstile-setup.md). - Human review on anomalies. If OCR misses a license number or the name on a certificate doesn't match the signup, the build still completes but the operator gets a Slack alert with a deep link to fix it.
You'll need an AWS account, a Stripe account, an Airtable base, and an Anthropic API key.
- Copy
.env.deploy.exampleto.env.deployand fill in your values. - Create the S3 bucket and wildcard DNS for your domain.
- Deploy the build Lambda:
./deploy.ps1 - Deploy the admin/signup Lambda: follow
ADMIN_SETUP.md(orADMIN_RUNBOOK.mdfor a from-scratch walkthrough). - Point a Stripe webhook (
checkout.session.completed) at the build Lambda's API Gateway URL.
Deeper internals: docs/DEVELOPER_GUIDE.md.
Not included: the marketing site and signup page (verifyanagent.com) live in a separate private repo.
Functional Source License 1.1 (Apache 2.0 future license). You can read, run, modify, and learn from this code for any purpose except offering a competing commercial product or service. Each release converts to Apache 2.0 two years after publication.
Built by Jude Wakim — cloud security engineer.
