ParallelCS Start here

HomeFoundationsWeek 6

Week 6 of 12

A neural net from scratch, no framework magic

Follow Karpathy micrograd, rebuild autograd from first principles, train a tiny MLP, explain backprop in your own words.

Why this resource? 60 second framing

Open the curated resource above and come back here when you have run it or watched it once. The coach is the right place to talk about what you saw.

What you ship by Sunday

Public repo micrograd-rebuilt with the student own from-blank reimplementation plus a notebook training a 2-layer MLP on sklearn make_moons with a printed loss curve. README has a 6-line backprop explanation in the student own words (Hinglish allowed).

The coach (Socratic, one hint per turn)

30 messages / day

The coach gives you one hint per turn. Never the answer. Hinglish bilkul welcome hai. It cannot do arithmetic; run code yourself and paste the result.

0 / 2000

Ask for a Socratic review of your artifact

Paste your README (or a draft of it). The coach will read it like an honest senior would, ask one question that surfaces the weakest link, and point at one thing to tighten. It will not rewrite it for you. Maximum 8000 characters.

0 / 8000

Before you close the tab

Three short questions to answer in your own words, even just in your head. The point is to notice what you actually learned this week, not to perform learning back at the coach.

  1. What broke first when you tried this week artifact, and what did you do?
  2. Pick one line of code that confused you initially. What does it do?
  3. If you had to rebuild this from a blank file tomorrow, what would you start with?