Multi-tenant RAG: isolating client data in AI agents
When you build an AI agent for one client, RAG is straightforward: embed their docs, store them, query them. But the moment you take a second client,…
// notes · build log
Real constraints, tradeoffs and things that broke while shipping AI, automation and web work — each one from a project that's live, written in Pedro's voice, not a content calendar.
AIWhen you build an AI agent for one client, RAG is straightforward: embed their docs, store them, query them. But the moment you take a second client,…
AutomationEvery AI image model I've used mangles text. Quotes, numbers, names — they come out as gibberish far more often than not. For many uses that's fine,…
AIA dentist I work with spends half the consultation typing notes into a patient record. The other half, they're wrestling with the UI. The gap between…
AILLM outputs are nondeterministic. The same prompt gives you a different answer every time. So how do you test a system whose core is a moving target?…
SEOWhen a client's real estate site has 1,400 listings in four languages and Google is showing exactly three of them, the problem isn't translation.…
AIThe first month of the Telegram bot was a lesson in model bills. Social Command Center turns one idea into four platforms of copy plus an AI image,…
WebWhen Cabarete Villas asked me to build their booking platform, they had a specific pain: guests booking the same villa on Airbnb and VRBO because the…
// hit one of these?
Every note comes from a system that's running. If you're stuck on the same problem, here's what the build looks like: