About

Xalpaca is a small studio of engineers who spent too long watching AI demos work beautifully on a laptop and fall apart on the second day of real traffic.

What we believe

• Boring beats clever. A queue, a timeout and a retry budget will outperform a novel architecture most weeks.
• Measure before you tune. Most “model quality” complaints turn out to be retrieval or preprocessing bugs.
• Hand it over. Every engagement ends with your team able to run the thing without us. If they cannot, we did not finish.

How we work

We keep the roster short on purpose — usually two or three active clients. That lets us stay in the same week as the people doing the work, instead of parachuting in for status meetings.

We write things down. Design decisions, incidents and the reasoning behind them live in files your team can read six months from now, not in a Slack thread.

Where we are

Distributed across a few time zones, with most overlap in the CET and US Eastern working hours. We work remotely and asynchronously by default.