Shipping LLM Features Without Burning the Budget
Prototype to production for AI features: streaming, cost control, prompt injection, and the failure modes that only appear under real traffic.
Writing
Long-form notes on problems I have actually had to solve — scaling real-time systems, keeping large Laravel codebases maintainable, Core Web Vitals, and shipping AI features that survive production.
Prototype to production for AI features: streaming, cost control, prompt injection, and the failure modes that only appear under real traffic.
Lighthouse scores 100 locally and your field data says otherwise. Here is what actually moves LCP, INP and CLS on a real Next.js App Router site.
A single Socket.IO process handles thousands of connections fine. Here is what actually breaks when you add the second one, and how to fix it.
Microservices are not the cure for a messy Laravel codebase. Module boundaries are. Here is the structure I use on long-lived projects.