Cloud & Architecture·3 min read·April 10, 2025

A full AWS and Postgres build, end to end

Supermatches.in was not a one-week MVP. It was a full build on AWS with Postgres at the core - the other end of the spectrum from managed-everything, and a good lesson in when that trade is worth making.

We talk a lot about one-week MVPs on managed platforms, because they are genuinely the right call surprisingly often. Supermatches.in is the useful counterexample — a full build on a complete AWS stack with Postgres at the core. Different project, different constraints, deliberately different decisions.

Why not just managed-everything

The honest answer to "managed platform or full AWS" is "it depends", and the things it depends on are worth being specific about. For supermatches.in, the build called for control over the pieces that a fully managed setup abstracts away — how the infrastructure was shaped, how data moved through it, how the parts fit together — and a managed platform that makes the easy things easy also makes those specific things harder.

So this was a full AWS build. More setup, more responsibility, more that we own. In exchange, no boundaries we did not choose. That trade is wrong for an MVP and right for a build where the architecture itself is part of the product.

Postgres at the core, again

Whether it is a one-week Supabase MVP or a full AWS build, the database underneath is Postgres. That is not a coincidence and it is not a default we have never examined — it is a deliberate, repeated choice.

Postgres is boring in the best possible sense. It is reliable, it is well understood, it does far more than people remember it does, and it scales much further than most projects will ever need. On supermatches.in it is the part of the stack we worried about least, which is exactly the property you want from the component that holds all your data. The interesting decisions in this build were everywhere else; the database just worked, which freed us to spend our attention on the parts that needed it.

The pitfalls of a full build

You now own the boring parts. Backups, security groups, networking, scaling, monitoring, the patching, the certificates — on a managed platform, much of this is someone else's job. On a full AWS build it is yours, and "we will set up monitoring later" is how you end up debugging a production issue with no visibility into what is happening. We treat the operational foundations as part of the build, not as a follow-up, because the follow-up version of that work happens during an incident.

Cost is now a design decision, continuously. A managed platform has a fairly legible price. A full AWS stack has a bill that is the sum of a hundred choices, and it is very easy to architect something that works fine and costs more than it should. We size deliberately and revisit it, rather than provisioning generously and discovering the consequence on the invoice.

Complexity is a tax you keep paying. Every additional service is another thing to operate, secure, monitor and reason about. The discipline on a full build is to add a piece of infrastructure only when there is a concrete reason, not because it might be useful. The AWS console makes it very easy to assemble more architecture than the problem actually has.

What we would tell you

The choice between a managed platform and a full cloud build is not about which is more serious. It is about what the project actually needs control over.

  • Default to managed when speed matters and the architecture is not the interesting part. Most early products are here.
  • Go full-stack on the cloud when you need control of the pieces a platform abstracts away — and accept that you are taking on the operational weight that comes with it.
  • Keep Postgres at the core either way. It is the decision you will never regret and rarely have to revisit.
  • Build the operational foundations in, not afterward. The "afterward" version is always written under pressure.

Supermatches.in did not need to ship in a week, and it did need control a managed platform does not give you. So we made the other trade. The skill is not preferring one approach — it is being honest about which one the specific project in front of you is asking for.

AWSPostgreSQLArchitectureCloudBackend

Working on something similar?

We have probably hit the exact scaling or architectural bottlenecks you are navigating. Book a call and we'll walk through the technical trade-offs.