Corey Kuwanoe ckuwanoe

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.

rclone systemd service

Preparation

This service will use the same remote name you specified when using rclone config create. If you haven't done that yet, do so now.

Next, create the mountpoint for your remote. The service uses the location ~/mnt/<remote> by default.

mkdir ~/mnt/dropbox

THIS IS OUTDATED. PLEASE FOLLOW THE FOLLOWING LINK

--> A beginner's guide to nflfastR <--

Basic nflscrapR tutorial

I get a lot of questions about how to get nflscrapR up and running. This guide is intended to help new users build interesting tables or charts from the ground up, taking the raw nflscrapR data.

Quick word if you're new to programming: all of this is happening in R. Obviously, you need to install R on your computer to do any of this. Make sure you save what you're doing in a script (in R, File --> New script) so you can save your work and run multiple lines of code at once. To run code from a script, highlight what you want, right click, and select Run line. As you go through your R journey, you might get stuck and have to google a bunch of things, but that's totally okay and normal. That's how I wrote this thing!

ESPN's hidden API endpoints

Football

College Football

Latest News: http://site.api.espn.com/apis/site/v2/sports/football/college-football/news

Latest Scores: http://site.api.espn.com/apis/site/v2/sports/football/college-football/scoreboard

	"""
	The most atomic way to train and run inference for a GPT in pure, dependency-free Python.
	This file is the complete algorithm.
	Everything else is just efficiency.

	@karpathy
	"""

	import os # os.path.exists
	import math # math.log, math.exp

	require "selenium/webdriver"

	Capybara.register_driver :chrome do \|app\|
	Capybara::Selenium::Driver.new(app, browser: :chrome)
	end

	Capybara.register_driver :headless_chrome do \|app\|
	capabilities = Selenium::WebDriver::Remote::Capabilities.chrome(
	chromeOptions: { args: %w(headless disable-gpu) }
	)

	#!/bin/bash

	# Parse df selected output
	df -h\|egrep -v 'File\|tmpfs\|docker\|udev'\| \
	while read LINE; do
	USED_NUMBER=`echo $LINE \|awk '{print $5}'\|sed 's/\%//'\|sed 's/ //g'`
	USED_PERCENT=`echo $LINE \|awk '{print $5}'\|sed 's/ //g'`
	MOUNT_POINT=`echo $LINE \|awk '{print $6}'\|sed 's/ //g'`
	if [ $USED_NUMBER -gt 80 ]; then
	# Create message without spaces

	export default class Ajax {
	get(url, callback) {
	let xhr = window.XMLHttpRequest ? new XMLHttpRequest() : new ActiveXObject('Microsoft.XMLHTTP');
	xhr.open('GET', url);
	xhr.onreadystatechange = () => {
	if (xhr.readyState > 3 && xhr.status === 200) {
	callback(xhr.responseText);
	}
	};
	xhr.setRequestHeader('X-Requested-With', 'XMLHttpRequest');

	require 'base64'
	require 'open-uri'
	require 'net/http'
	require 'net/https'
	require 'json'

	class OCR
	attr_reader :api_key, :image_url

	def self.scan(api_key:, image_url:)