Remotion Lambda Migration: Cost, Quota Math, skip-worktree
Migrating Remotion from local rendering to AWS Lambda — the quota trap (10, not 1000), real cost at $15/video, and the git skip-worktree fix.
Nonce-Based CSP in Next.js 16: Middleware to proxy.ts
Implementing nonce-based Content Security Policy in Next.js 16 using proxy.ts instead of middleware.ts. Full code, CSP directives, common mistakes.
Claude Code Setup Guide 2026: 20 Agents, 8 Skills, 11 Hooks
Complete Claude Code OS: 20 custom agents, 8 slash-command skills, 11 automated hooks, 5 aliases, 50-file memory system. Includes 30-minute starter kit.
Supabase RLS Gotchas: Triggers, FK SET NULL, Migrations
Three non-obvious Supabase RLS gotchas — SECURITY DEFINER triggers, FK SET NULL needing UPDATE policies, and partial migrations — with SQL fixes for each.
Anatomy of 4 Stacked Bugs: Production Incident Debugging
How 4 stacked bugs in a tax-tech system masked each other across 2 days and 3 commits. Real timeline, root causes, debugging playbook for compounding failures.
IDOR Prevention Checklist for Next.js + Supabase APIs
An IDOR slipped past code review during a billing rewrite. Here is the 6-point checklist and defense-in-depth approach I now use on every pull request.
AES-256-GCM Encryption for PII in Node.js — 4 Gotchas
How to encrypt SSNs and PII at rest with AES-256-GCM in TypeScript. Covers key rotation traps, IV reuse, auth tags, and a full production implementation.
Nightly QA System at 2 AM: What It Caught That CI Missed
A 15-group nightly QA system running 35 test files at 2 AM catches integration bugs CI never will. Real examples, architecture, and $0.50/night ROI breakdown.
Extraction QA Loop: Turn Failed Runs Into Training Data
An extraction QA loop turns failed runs into labeled training data, lifting accuracy from 87% to 95% in 6 weeks with zero manual labeling.
Azure Content Understanding vs Document Intelligence
After testing Azure CU and DI on 50 financial documents, CU hit 95% accuracy vs DI's 87%. DI misreads form titles as values. Full comparison with code.
MCP Server Migration: Replacing an In-House Tax Engine
How a tax-tech app replaced a brittle in-house calculation engine with an MCP server. Benchmarks (2.2s), a silent percentage bug, PII anonymization.
Transactional Email Batching With Redis SETNX
Redis SETNX with a TTL and a digest endpoint batches transactional emails so users get one notification, not thirteen. Code, architecture, real incident.