Mikhail Podgurskiy kmmbvnr

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.

WSGI and ASGI on a single server

We want to attach a an ASGI app, like FastAPI, to an existing Django app, without any performance downsides.

In our example we are attaching an MCP server, including auto-loading.

Preparation

You have an existing Django app, you serve via WSGI on a server like guinicorn. That's it!

	"""
	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

	"use client";


	// Import necessary utilities and components.
	// tailwind-merge is used to intelligently merge Tailwind CSS classes, preventing conflicts.
	import React from "react";
	import { twMerge } from "tailwind-merge";
	// useState is a React Hook for managing state within the component.
	import { useState } from "react";
	// Custom icon components for the sidebar.

	import React from "react";

	// yea
	import axios from "axios";
	import {render} from "react-dom";
	import {setStylesTarget} from "typestyle";

	import {Provider} from "reactivated/context";
	// nope

	<div #surfaceAnchor mdcMenuSurfaceAnchor (click)="onClick($event)">
	<ng-content (keydown)="onKeydown($event)" select="mdc-text-field"></ng-content>
	<ng-content select="mdc-menu"></ng-content>
	</div>


	$fn=60;

	// Answer from: https://stackoverflow.com/users/1320888/cutetare
	// https://stackoverflow.com/questions/33146741/way-to-round-edges-of-objects-openscad/33289349#33289349
	module roundedcube(xx, yy, height, radius) {

	difference(){

	cube([xx,yy,height]);

	# Stolen from: https://medium.com/survata-engineering-blog/monitoring-memory-usage-of-a-running-python-program-49f027e3d1ba

	print("\33c")

	import tracemalloc


	def measure_memory(func):
	tracemalloc.start()

	"""
	This is an extended HTTP response class for Django >=2.0 that adds lots of fancy features.

	It's most useful when needing to accellerate slow view functions that return templates,
	but it can be used anywhere where you need to return an HTTPResponse() or render(...).

	It works by subclassing the builtin Django HttpResponse and StreamingHttpResponse,
	and adding improvements in many areas, including functionality, speed, and security.

	The most up-to-date version of this code can be found here:

	brew install caddy mkcert nss dnsmasq

	mkcert -install

	mkcert '.app.test' '.cdn.test'

	# rename the certs and move them under /usr/local/etc/caddy/certs

	cat <<EOF > /usr/local/etc/caddy/Caddyfile
	.app.test:443, .cdn.test:443 {