Notes from Production

A no-look touch-typing trainer that floats an on-screen keyboard over any app and mirrors your real typing in real time. The threading rule that keeps it stable, and shipping it to a public release in one session.

Building a self-running HubSpot enrichment with n8n, SerpAPI, and Claude. How I wired deep web search and an AI agent to auto-populate music labels into hundreds of artist contacts.

What I learned building a personal desktop widget that syncs realtime with ClickUp. Three rounds for a complete sync. Eleven rounds for a small UI element. The straight line is fiction.

A working-method walk-through of QoreX: five Claude-powered agents with names, roles, handoffs, and a shared brain that ships real production work.

Building the AutomaQue CRM chatbot, I skipped embeddings entirely. For personal-scale data, the right answer is to pass the whole dataset as context - and the upgrade path is still wide open.

Building QolAssist taught me what 'live' actually means when speech transcription and Claude tokens have to share an event loop without blocking each other.

The difference between a workflow that runs and one that survives. Idempotency, backoff, state persistence, dev/prod separation.

What the marketplace docs leave out. Practical guide with code snippets that actually work in production.

The BIGSELLER story - how I turned a closed platform into an open one with DevTools, cookies, and a polling state machine.

It's not ChatGPT wrappers. Here's what I'm actually building in production - and why the hard part is the boring part.