Mike Wills MikeWills

LLM Wiki

A pattern for building personal knowledge bases using LLMs.

This is an idea file, it is designed to be copy pasted to your own LLM Agent (e.g. OpenAI Codex, Claude Code, OpenCode / Pi, or etc.). Its goal is to communicate the high level idea, but your agent will build out the specifics in collaboration with you.

The core idea

Most people's experience with LLMs and documents looks like RAG: you upload a collection of files, the LLM retrieves relevant chunks at query time, and generates an answer. This works, but the LLM is rediscovering knowledge from scratch on every question. There's no accumulation. Ask a subtle question that requires synthesizing five documents, and the LLM has to find and piece together the relevant fragments every time. Nothing is built up. NotebookLM, ChatGPT file uploads, and most RAG systems work this way.

AI Time Management System — Claude Project instructions for daily planning, habit tracking, and weekly reviews using Todoist + Google Calendar

AI Time Management System — Claude Project Instructions

What this is: Instructions for a Claude Project that turns Claude into a personal time management assistant using Todoist and Google Calendar. Paste this into a Claude Project's custom instructions, connect Todoist and Google Calendar via MCP, and you get an AI assistant that plans your day, schedules tasks, tracks habits, and runs daily/weekly reviews.

Requirements: Claude Pro/Team with a Claude Project, Todoist MCP connector, Google Calendar MCP connector.

How to customize: Search for [CUSTOMIZE] tags — these mark the sections you need to adapt to your own life. The system design works as-is; you just need to fill in your details.

A Very Short Guide to Writing Guides

This is just a few thoughts on the topic of writing technical guides. This was intended for Basho's engineering team, but this may apply to open source projects in general.

Audience

It's commonly preached that the first step in writing is to identify your audience; to whom are you writing? This is the most well known, most repeated, and most overlooked step of writing in general and technical writing in particular. Take this document, for example. My audience is technical people who need to communicate technical information, and not teenagers, so I shy away from images of pop icons and memes. I use jargon and words like "identify" rather than "peep this".

	<style>
	.thumbnail.with-caption {
	display: inline-block;
	background: #f5f5f5;
	}
	.thumbnail.with-caption p {
	margin: 0;
	padding-top: 0.5em;
	}
	.thumbnail.with-caption small:before {

	%w(rubygems sequel fileutils yaml active_support/inflector).each{\|g\| require g}

	require File.join(File.dirname(__FILE__), "downmark_it")

	module WordPress
	def self.import(database, user, password, table_prefix = "wp", host = 'localhost')
	db = Sequel.mysql(database, :user => user, :password => password, :host => host, :encoding => 'utf8')

	%w(_posts _drafts images/posts/featured).each{\|folder\| FileUtils.mkdir_p folder}

	[
	{"US":"United States"},
	{"CA":"Canada"},
	{"AF":"Afghanistan"},
	{"AL":"Albania"},
	{"DZ":"Algeria"},
	{"DS":"American Samoa"},
	{"AD":"Andorra"},
	{"AO":"Angola"},
	{"AI":"Anguilla"},


	window.Countries = {};
	$.ajax({
	url: "http://where.yahooapis.com/v1/countries",
	data: {appid: "5mCqm._V34HyJ7CniyiCQq.oi8LjlbUnJ1ge4X36P9bpyyao2ey7xvS.mBb1jcAf69AzoQ--", format: "json"},
	dataType: "json",
	success: function(data){
	var countries = _.map(data.places.place, function(country){
	return {"name": country.name, "url": country.uri};
	})