Engineering
Perceived speed is real speed. How we shaved 400ms off every generation by rethinking the loading state.
Feb 10, 2026 · 9 min read
Users judge latency before they read words. If the interface feels idle, they assume the system is slow even when backend timings are acceptable.
We replaced blocking loaders with immediate skeleton context and progressive text streaming. That changed the interaction from waiting to reading.
The measurable gain was modest in raw milliseconds, but perceived responsiveness improved significantly in session retention and completion rates.
The difference between AI copy that converts and AI copy that gets ignored isn't word count — it's voice. Here's how we think about it.
Apr 2, 2026 · 6 min read
EngineeringA deep dive into the system behind Veloraz brand voices, and what we learned training models on 40,000+ writing samples.
Mar 18, 2026 · 11 min read
TeamShipping cadence is a craft. We break down the loop we use to publish something useful every Friday, for 48 weeks straight.
Mar 5, 2026 · 8 min read
