Mike Heavers heaversm

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.

JavaScript GUI libraries

These libraries can be used to quickly create a GUI for configureable parameters using sliders, checkboxes, colors pickers etc

Tweakpane https://github.com/cocopon/tweakpane Demo: https://cocopon.github.io/tweakpane/
control-panel https://github.com/freeman-lab/control-panel
ControlKit https://github.com/automat/controlkit.js
guify https://github.com/colejd/guify Main site is down, here is the demo https://jons.website/projects/guify/index
oui https://github.com/wearekuva/oui
Palette.js https://github.com/lehni/palette.js

Quick, Draw! ndjson data

The Quick, Draw! dataset uses ndjson as one of the formats to store its millions of drawings.

We can use the ndjons-cli utility to quickly create interesting subsets of this dataset.

The drawings (stroke data and associated metadata) are stored as one JSON object per line. e.g.:

	<!DOCTYPE html>
	<html>
	<head>
	<title>Speaker Authenticator</title>
	<script src="https://cdnjs.cloudflare.com/ajax/libs/recorderjs/0.1.0/recorder.js"></script>
	</head>

	<body>
	<div class="container">
	<h1>Validate Speaker</h1>

	import React from "react";
	import ReactDOM from "react-dom";

	import "./styles.css";

	class GithubCommit extends React.Component {
	constructor() {
	super();
	this.state = {
	commitHistory: [],

	package org.opencv.android;

	import java.util.List;

	import org.opencv.R;
	import org.opencv.core.Mat;
	import org.opencv.core.Size;

	import android.app.Activity;
	import android.app.AlertDialog;

	# repair luarocks
	sudo rm -rf ~/.cache/luarocks/
	luarocks install loadcaffe
	luarocks install cutorch
	luarocks install cudnn
	luarocks install cunn

	# get cudnn (https://developer.nvidia.com/cudnn) and then (assuming cudnn 7.*) ...
	git clone https://github.com/soumith/cudnn.torch.git -b R7 && cd cudnn.torch && luarocks make cudnn-scm-1.rockspec && cd .. && rm -rf cudnn.torch

	#!/usr/bin/env python3

	from PIL import Image
	import numpy as np
	import tensorflow as tf
	import tensorflow_hub as hub

	# smooth values from point a to point b.
	STEPS = 100
	pt_a = np.random.normal(size=(512))

	'use strict';

	const autoprefixer = require('autoprefixer');
	const ExtractTextPlugin = require('extract-text-webpack-plugin');
	const path = require('path');
	const {
	CheckerPlugin,
	TsConfigPathsPlugin,
	} = require('awesome-typescript-loader');
	const HappyPack = require('happypack');

	/* global gapi */

	const API_KEY = 'YOURAPIKEYHERE';

	import React, { Component } from 'react';

	class App extends Component {

	loadYoutubeApi() {
	const script = document.createElement("script");