Insights

Notes from the work itself

What we learned building the systems, written down while it was still fresh. No thought leadership — just the decisions, the trade-offs, and the parts that turned out to matter.

6 articles

AI & Data

Your RAG pipeline is retrieving the wrong chunks

Most retrieval failures are not model failures. They are chunking failures, and they show up as confident answers assembled from the wrong three paragraphs.

Aug 18, 20265 min read

Engineering

Row-level security is the tenant boundary you already have

Every multi-tenant app enforces isolation in application code until the day one query forgets. Postgres will hold that line for you, in the database, where it cannot be forgotten.

Aug 11, 20264 min read

Engineering

The compounding cost of a badly drawn service boundary

A boundary in the wrong place does not announce itself. It bills you slowly, in coordinated deploys and distributed transactions, until a feature that should take a week takes a quarter.

Jul 29, 20264 min read

AI & Data

Training the model is ten percent of the work

The notebook that hits 94% accuracy is the beginning. What separates a demo from a system is everything that happens after the metric looks good.

Jul 16, 20264 min read

Fintech

The ledger you wish you had built first

Storing a balance as a number on the account row is the decision every payments system regrets. Double-entry is older than software and still the correct answer.

Jul 2, 20264 min read

Cloud & DevOps

Cutting a forty-minute CI pipeline down to six

Slow CI is not a tooling problem. It is a queue that quietly changes how a team works, and most of the time it is fixable in a day without touching a single test.

Jun 24, 20264 min read

Let's build something that lasts

Tell us about your project and we'll get back to you within 12 hours with next steps.

Start a project See our work