July 1, 2026 · 5 min read
Static-first, AI-ready: the architecture behind ingenx.ai
This site is its own case study: S3 and CloudFront for the pages, Lambda and Bedrock for the conversation, Terraform for all of it — at a run cost under ten dollars a month.
We tell clients that architecture should be boring where it can be and intelligent where it must be. This website practices that: every page is pre-rendered and served from S3 behind CloudFront — nothing to patch, nothing to scale, and edge latency in Mumbai, Dubai and Riyadh alike. The only servers involved run for milliseconds at a time.
The dynamic seam is deliberate
Two API routes exist behind the same CloudFront distribution: a contact pipeline (Lambda → DynamoDB first, then SES and SNS — a lead is never lost to an email failure) and the assistant you can open on this page, a Lambda calling Bedrock with retrieval grounded strictly on our published content. Guardrails keep it from inventing prices or commitments; commercial questions go to humans.
Bilingual is structural, not cosmetic
English and Arabic are separate locale trees with full right-to-left layout, hreflang alternates, and equal content — the same standard we hold client work to. If your Arabic experience is a translation pass at the end, your Gulf customers can tell.
Everything — DNS, certificates, CDN, WAF, functions, budgets — is Terraform in a single repository, deployed by GitHub Actions with OIDC and no stored keys. Cloning the stack for a client is a variables file and a content swap. That repeatability is the product; the website is just its first customer.