How a 1.2 MB ONNX model and a few hundred lines of glue replaced a recurring cloud cost — and made the product faster and more private.
How a 1.2 MB ONNX model and a few hundred lines of glue replaced a recurring cloud cost — and made the product faster and more private.
How a 1.2 MB ONNX model and a few hundred lines of glue replaced a recurring cloud cost — and made the product faster and more private.
A vibe-coded prototype gets you to “it works on my machine.” The last stretch — auth, observability, deploys, the failure modes nobody demos — is where real products are won. Here is the checklist I run every time.
You do not need a Vault cluster to stop leaking credentials. A pragmatic setup for solo builders and small teams that scales when you do.
Naming, versioning, and error shapes are cheap to get right on day one and brutally expensive to fix later. A field guide.