Vedran Mandić vmandic

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.

	<!DOCTYPE html>
	<html lang="en">

	<head>
	<meta charset="UTF-8">
	<meta name="viewport" content="width=device-width, initial-scale=1.0">
	<title>My Angular from Scratch</title>
	<style>
	.my-component {
	font-family: Arial, sans-serif;

	using System;
	using System.IO;
	using System.Threading.Tasks;
	using Microsoft.AspNetCore.Hosting;
	using Microsoft.Extensions.Configuration;
	using Microsoft.Extensions.Hosting;

	namespace StartupIConfigurationInjectionProblem
	{
	public class Program

	[CmdletBinding()]
	param (
	[Parameter(Mandatory)]
	[string]
	$RiderHome
	)

	function Update-WindowsDefenderForRider {
	[CmdletBinding()]
	param (

	using Microsoft.AspNetCore.Hosting;
	using Microsoft.AspNetCore.Builder;
	using Microsoft.AspNetCore.Http;
	using Microsoft.Extensions.Hosting;

	public class Program
	{
	public static void Main(string[] args) =>
	Host.CreateDefaultBuilder(args)
	.ConfigureWebHostDefaults(webBuilder =>

	# dotnet --list-sdks # output looks like
	# 2.1.500 [/usr/local/share/dotnet/sdk]
	# awk '{print $2"/"$1}' # takes each line from stdin and prints out the 2nd column then "/" then the first column
	# this gives us [/usr/local/share/dotnet/sdk]/2.1.500
	# sed 's/\]//' # replaces the first ']'. ']' has to be escaped with \
	# sed 's/\[//' # replaces the first '['. '[' has to be escaped with \
	# grep -Ev '2\.2\.102' # matches on 2.2.102 which was the latest SDK installed
	# -E # enables extended regular expression syntax
	# -v # inverts the match. So return anything that doesn't match
	# xargs # allows us to pass standard in as an argument to rm -rf

	/*

	* Publish/Subscribe global events functions - implementation of pubsub pattern for loose coupled components *

	Usage:

	* Publish global application event:

	publish("unique event name / category / subcategory", arg1, arg2, arg3);
	or

	using System;
	using System.Threading.Tasks;
	using NUnit.Framework;

	namespace PuppeteerSharp.Contrib.Sample
	{
	public class Examples
	{
	async Task<IBrowser> Browser()
	{

	USE [HobbyDB]
	GO
	/**** Object: Table [dbo].[AspNetRoleClaims] Script Date: 6/4/2018 10:18:03 PM ****/
	SET ANSI_NULLS ON
	GO
	SET QUOTED_IDENTIFIER ON
	GO
	IF NOT EXISTS (SELECT * FROM sys.objects WHERE object_id = OBJECT_ID(N'[dbo].[AspNetRoleClaims]') AND type in (N'U'))
	BEGIN
	CREATE TABLE [dbo].[AspNetRoleClaims](