Simor
Data Infrastructure for Production AI
Practical writing on AI data engineering, feature stores, and the infrastructure choices that determine whether AI systems work in production.
Large language models suffer from a critical flaw: their knowledge is frozen at training time, encoded implicitly in billions of parameters, and prone to confident fabrication. This limitation becomes
You're at a vending machine, desperately needing caffeine. You insert a dollar, press B4 for coffee, but nothing happens. Did the machine eat your money? Did it register the button press? In frustrati
Large language models shattered the deterministic paradigm of traditional software. The same prompt can produce different outputs. Model behavior emerges from billions of parameters trained on vast te
Picture yourself at a busy airport gate. The agent announces: "We'll now board passengers in rows 20 through 30." Simple, efficient, everyone knows whether it's their turn. Now imagine instead they sa
A nightclub bouncer with a peculiar condition: they never forget a face they've seen, but sometimes they think they've seen faces they haven't. When someone approaches, they'll either say "You've defi
AI systems increasingly make decisions that profoundly affect human lives. Healthcare systems deny treatment recommendations based on zip codes. Hiring platforms filter resumes based on gender. Crimin
Russian nesting dolls (Matryoshka) are wooden dolls where each one opens to reveal a smaller doll inside, which opens to reveal another, and so on. Each doll represents an operation in your distribute
AI systems introduce attack vectors that don't exist in traditional software. Unlike conventional applications that process data according to fixed rules, AI models learn from data, making them vulner
A postal service where every postcard has a strict template. The address fields are always in the same spot. The message area has specific sections for specific types of information. Both sender and r