Writing
Notes on building software: backend, integrations and applied AI. Problems I ran into, decisions I made and what I learned along the way.
Subscribe via RSSDB-Fat, LLM-Light — or how I stopped asking the AI to decide things
The model can propose an appointment. The backend still has to check availability and save it. How I separate conversation from the rules that control a booking.
The fourth time I rebuilt the same state machine, I made it a library
I built the same FSM four times for four different projects. The fourth version became reactive-fsm — a zero-dependency library for controlling LLM tool access in production.
The tax authority doesn't care that you were following SOP
I deleted a user from AD at 18. Almost cost a company its tax audit trail. The problem wasn't the delete — it was that nobody in the room understood what the data meant.
Network Asymmetry: Why Your App Needs a Resilience Layer
External APIs fail. Exponential backoff, circuit breakers, and fallback behavior for a clinic's booking system.
Server-Centric State: Authentication Patterns for Next.js
Critical state belongs on the server. Examining HttpOnly cookies, server-side sessions, and zero client-side token storage.
Deterministic AI: Validating LLM Output in Production
LLMs are non-deterministic by nature. How to wrap every model output with Zod schemas and finite state machines to guarantee stability.