Parmesh Krishen therealparmesh

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.

The Reality of "Just Render HTML"

A complete breakdown of the architecture, tooling, and paradigms involved when returning to a server-driven architecture using HTMX, Tailwind, and Vanilla JS. This highlights how drastically the frontend footprint shrinks when the server reclaims state and routing.

Core Architecture & Language

Hypermedia: HTMX (HTML attributes, out-of-band swaps, lazy-loading partials)
Language: Vanilla JS
Component Paradigms: Server-side templating (partials, fragments, includes), standard HTML elements

The Reality of "Just a React App"

A complete breakdown of the architecture, tooling, and paradigms involved in modern frontend development. This highlights the true scope of work required when React is the default (and often only) option for frontend developers who are effectively React specialists.

Core Architecture & Language

React: Virtual DOM, JSX
Component Paradigms: Class component lifecycles, Higher-Order Components (HOCs), render props, Hooks
Advanced Rendering: Suspense, Error Boundaries

Zed Vim Mode Cheat Sheet

Zed's Vim mode replicates familiar Vim behavior while integrating modern features like semantic navigation and multiple cursors. This cheat sheet summarizes essential shortcuts and settings to help you navigate and edit code efficiently in Zed.

Enabling/Disabling Vim Mode

Enable/Disable Vim Mode: Open the command palette and use Toggle Vim Mode.
This updates your user settings: "vim_mode": true or false.

	autoload -Uz compinit
	compinit

	setopt append_history
	setopt auto_pushd
	setopt autocd
	setopt cdable_vars
	setopt complete_in_word
	setopt correct
	setopt glob_dots

	bunny() {
	local script
	script=$(bun run \| sed 's/\x1b\[[0-9;]*m//g' \| rg --replace '$2' '(\s+)\$ bun run (\S+)' \| fzf)
	if [[ -n "$script" ]]; then
	bun run "$script"
	fi
	}

	(async () => {
	try {
	const {
	sheet: { cssRules },
	} = document.getElementById('react-native-stylesheet');
	const cssText = Array.from(cssRules)
	.map(({ cssText }) => cssText)
	.filter(Boolean)
	.join('\n\n');
	console.log('--- Plain CSS Output ---');

	export function html(strings: TemplateStringsArray, ...expressions: string[]) {
	let result = '';
	for (let i = 0; i < strings.length; i++) {
	result += strings[i];
	if (i < expressions.length) {
	result += expressions[i];
	}
	}
	return result;
	}

	<!doctype html>
	<html lang="en">
	<head>
	<meta charset="utf-8" />
	<meta
	name="viewport"
	content="width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no"
	/>
	<title>reveal.js</title>
	<link

	export async function retryAsync<T>(
	asyncFn: () => Promise<T>,
	maxRetries: number,
	): Promise<T> {
	let retries = 0;

	while (retries < maxRetries) {
	try {
	return await asyncFn();
	} catch (error) {