Umut Keltek umutkeltek

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.

	# I use MacOS VMs from github for iOS development.
	# By no suprise they are a bit slow and have a lot of things I don't use
	# Hence this script for lighter and better VM for my iOS development and builds
	# GUI and animation related things to tweak
	defaults write NSGlobalDomain NSAutomaticWindowAnimationsEnabled -bool false
	defaults write NSGlobalDomain NSWindowResizeTime -float 0.001
	defaults write -g QLPanelAnimationDuration -float 0
	defaults write com.apple.dock autohide-time-modifier -float 0
	defaults write com.apple.dock launchanim -bool false
	sudo sysctl debug.lowpri_throttle_enabled=0

	from deepspeed.launcher.runner import main
	import sys
	import os

	if sys.platform == "win32":
	os.environ["PL_TORCH_DISTRIBUTED_BACKEND"] = "gloo"

	if __name__ == '__main__':
	main()

	# .cursorrules

	Components & Naming

	- Use functional components with `"use client"` if needed.
	- Name in PascalCase under `src/components/`.
	- Keep them small, typed with interfaces.
	- Use Tailwind for common UI components like textarea, button, etc. Never use radix or shadcn.

	Prisma


	<TEMPLATE>
	<INSTRUCTIONS>
	Use the <CODEBASE> code as reference, and convert the high-level <TASK> into a set of very detailed step-by-step instructions that an AI coding agent can complete.
	Only includes steps an AI coding agent can take. Do not include testing or any other work a human would do to confirm the task has been completed.
	ALWAYS have the agent run a build when it is complete. Be specific and decisive about what the agent should do.
	Do not include any additional meta instructions to the user. Use markdown formatting.
	</INSTRUCTIONS>
	<TASK>

	"""
	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

	// Unity C# Cheat Sheet
	// I made these examples for students with prior exerience working with C# and Unity.
	// Too much? Try Unity's very good tutorials to get up to speed: https://unity3d.com/learn/tutorials/topics/scripting