Insights

Notes from the work

Practical writing on building software, shipping AI features, integrating APIs and running systems in production.

Laravel 13 puts embeddings, reranking and vector similarity in the query builder. For a good share of AI work, the separate Python service h...

A headless browser is the most expensive way to fetch a page, and it is the first thing most scrapers reach for. The gap between that and a...

The demo works and the production system does not. Almost every time, the failure is in retrieval rather than the model — and it is fixable.

Verification accuracy is mostly decided before the model runs. Capture quality, retries and review queues matter more than the matching algo...

Most applications are deployed onto infrastructure sized for a scale they will never reach. Here is how we decide what a system actually nee...

Every integration eventually sends the same request twice. Whether that is harmless or expensive is decided by design, not by luck.

An agent that only answers questions is a chatbot. Once it can act, the engineering problem shifts from accuracy to authority.

Websites are rebuilt every few years and the content outlives every one of them. Model it so the next redesign is a template change.

Building something we should be talking about?

If a problem in one of these articles looks like yours, we would like to hear about it.

Start a Project Explore Our Work