Engineering log

Engineering deep-dives, written from work we just shipped.

One long-form post per month on the things we ship for clients: multi-tenant SaaS architecture, AWS platform discipline, productized ECS deployment, internal platform engineering with AI in the loop, and LLM features that don't fabricate.

Every post comes out of work we just shipped on one of the production SaaS apps we run on the platform — Comms, ExpertCV, PilatesTime, RuralOps.

Posts

Three posts shipping in the next two months. The schedule drops to roughly every three weeks after.

  1. Building a single API for Email and WhatsApp without becoming a CPaaS

    Email through Resend, WhatsApp through a separate BSP. Two SDKs, two bills, two ways delivery breaks. We collapsed it into one contract without becoming a CPaaS. The dual-store decision — Postgres for facts, DynamoDB for the message hot path — is the part most teams get wrong.

    Read the post
  2. Every production SaaS app we run. One ECS cluster. One ALB. Here's how.

    Heroku is too expensive. Kubernetes is too heavy. ECS Fargate is where most 10–40-engineer teams land, but the public reference architectures are thin. This is a real one — shared ALB with host-based routing, the GitHub OIDC deploy role, the alarm pack that catches what AWS-native monitoring misses, and the cost economics at the cluster shape we actually run.

  3. Productizing platform engineering: from one-off Terraform to a service line

    Most platform teams have a Slack channel and tribal knowledge. We productized ours into a five-skill Claude Code plugin — aws-readonly, aws-infra, ci-pipeline, knowledge-base, platform-allowlist — backed by a cross-account AI_AGENT_READONLY IAM role and an allowed-commands.yaml registry that compiles to native permissions. Different category from Backstage / Port.