#productengineering

The demo in Turistar signs visitors into a real, shared account. They can move activities, edit a...

A single-presenter 'animated' feature meets a real multi-person cast: one twin stylized and two left photoreal in the same frame, a redeploy that silently re-buys an expensive identity render, and a planner that finally declares which cast member a shot is about.

A few months ago, a small Tailwind config change broke position: sticky across our nav and header...

If an agent can create unlimited worlds, characters, and images, a product quickly faces two...

Introdução: A Engenharia como Criatura Liminar e Colaborativa O maior erro do engenheiro...

For decades, software products were built around explicit instructions. A user clicked a button. The...

Moving a project tracker's hardcoded lists into Supabase-backed admin tables, shipping an Employee Management and Evaluations suite with access enforced at the API layer, and closing an SSRF-adjacent credential leak in a shared proxy.

Turning 'animated' from a reference-derived accident into a deliberate operator choice in an AI ad pipeline — and the identity-anchor, face-check, and negative-prompt bugs that only existed once animated people became a real code path.

Collapsing four separate ad-creation entry points into one brief-driven flow, building a VSL Generator and an AI Presenter mode on top of it, and the sample-rendering and cost-quoting accuracy work needed to make the pre-review sample trustworthy before a paid run.

Every plan-quality check in an AI ad pipeline had quietly assumed a reference video existed to measure against. Building a script-only mode meant finding every one of those assumptions the hard way — by running it and paying for the answer.

Our algorithm decided a driver had travelled 4km less than they thought. It was probably right....

An LLM provider is the service that runs the language model behind your product's AI features. The...