A notebook of what we built and why, written after each session with Claude Code. Each lesson explains the ideas in plain English, shows how it was done, what went wrong, and ends with a short quiz and something to try yourself.
Finding big files, spotting true duplicates with hashes, organising by type, and why "never delete without asking, always keep a backup" became the house rule.
Installing a modern Python, why virtual environments matter, and the detective story of a Jupyter kernel that kept dying — solved with a tiny C library.
Building a lottery-results dashboard (HTML + a small Python server + Excel), and why the honest feature — a backtest against random picks — matters more than any "prediction".
How "skills" give Claude a specialist's playbook, and what the canvas-design and frontend-design skills taught us about avoiding generic, AI-looking pages.
How we made an AI assistant that answers only from your own FAQ — embeddings, chunking, cosine similarity, and why it doesn't make things up.
Embeddings are just numbers. When a plain file is enough, when a vector database earns its place, and the exact-vs-approximate trade-off.
Turning PDFs, Word files and web pages into knowledge bases — text extraction, chunking with overlap, exports, and an offline file that works without internet.
Uploading over encrypted FTP, password-protecting pages with .htaccess, locking private folders, backing up before every change, and proving it all with tests.
How passwords are stored safely, how login tokens work, why Pro checks must happen on the server, Stripe for payments, and designing so you can migrate later.
House rules, memory, confirming before acting, backups, and the debugging habits that kept recurring across every project.