
Snack-size RAG - Guardrail Garage #003
A small, practical plan for building a snack-size RAG assistant with a tiny local LLM on a VPS. Goals, constraints, and a quick tour of RAG options, plus a few personal updates.

A small, practical plan for building a snack-size RAG assistant with a tiny local LLM on a VPS. Goals, constraints, and a quick tour of RAG options, plus a few personal updates.

A hands-on introduction to Claude Code as an AI pair programmer, with setup tips, workflow patterns, and hard-won lessons from a two-week refactor sprint. Includes guidance on planning, testing, and keeping control over agentic tools.

A practical introduction to getting better results from LLMs: free ways to compare models, meta-prompting, a simple prompting framework, and Socratic questioning. Part of the Guardrail Garage series.

A practical tour of tokenisation for LLMs. Why word-level fails, how subwords and BPE work, and what modern tokenisers look like, with tips and resources. Plus a brief personal update.

An accessible tour of word embeddings, from Word2Vec to GloVe, with intuition, a touch of maths, and practical notes on training and using them. Includes caveats, examples, and where modern models fit.

Do machines understand words? And if not exactly, do they use Google Translate?