Writing
Notes from
production.
Field notes on software architecture, APIs, payments, cloud and AI — from production, not theory.
Jun 26, 2026
From Monolith to Microservices to AI Features
Why a clean service boundary is the real prerequisite for adding AI to an existing product.
Read →
Jun 26, 2026
Cost-Aware LLM Inference on AWS
Practical ways to keep an AI feature’s cloud bill from quietly outgrowing its value.
Read →
Jun 26, 2026
Zero-Downtime Deploys for AI-Powered Services
Shipping continuously when your service also loads models and serves inference.
Read →
Jun 26, 2026
Designing REST APIs That AI Agents Can Use
The same traits that make an API pleasant for humans make it usable for autonomous agents.
Read →
Jun 26, 2026
Angular at 700K Users: Performance That Scales
The frontend habits that kept a high-traffic rewards platform fast under real load.
Read →
Jun 26, 2026
WhatsApp Assistants That Don’t Annoy People
Lessons from building WhatsApp automations that citizens and customers actually want to use.
Read →
Jun 26, 2026
Automating Back-Office Work with n8n and an LLM
Where a workflow tool plus a single AI step quietly removes hours of manual operations.
Read →
Jun 26, 2026
RAG on PostgreSQL with pgvector
You probably don’t need a new vector database. The Postgres you already run can do retrieval well.
Read →
Jun 26, 2026
Adding an AI Assistant Without Breaking Payments
How to bolt a helpful LLM onto a loyalty or checkout flow without putting money paths at risk.
Read →
Jun 26, 2026
Shipping APIs in the Age of AI Copilots
AI assistants make writing endpoints faster — they don’t make your contracts, security or load tests optional.
Read →
Jun 24, 2026
Integrating Multiple Payment Gateways Across Five Countries
Lessons from building a single checkout that handles multi-currency, POS routing, and reconciliation.
Read →
Jun 24, 2026
Zero-Downtime Deploys with GitHub Actions and Nginx
How I ship to production during business hours without a maintenance window.
Read →
Jun 24, 2026
Designing APIs That Survive Production Load
Versioning, idempotency, and the small decisions that keep an API stable when traffic spikes.
Read →