micrograd

name	micrograd
description	Interactive educational autograd engine — port of Karpathy's micrograd. Build, train, and visualise tiny neural networks step-by-step to learn how backpropagation works.
version	0.1.0
publisher	acmecorp
handler_language	typescript
handler_file	handler.ts
inputs_schema	{"type":"object","properties":{"action":{"type":"string","enum":["create_session","destroy_session","list_sessions","get_walkthrough_step","advance_walkthrough","create_model","train_model","train_with_snapshots","get_decision_boundary","generate_moon_dataset","get_stats"]},"session_id":{"type":"string"},"org_id":{"type":"string"},"step_index":{"type":"number"},"architecture":{"type":"array","items":{"type":"number"}},"dataset_type":{"type":"string","enum":["xor","moon","custom"]},"learning_rate":{"type":"number"},"epochs":{"type":"number"},"loss_function":{"type":"string","enum":["mse","hinge","bce"]},"seed":{"type":"number"},"snapshot_every":{"type":"number"},"custom_data":{"type":"object","properties":{"xs":{"type":"array"},"ys":{"type":"array"}}}},"required":["action"]}
outputs_schema	{"type":"object","properties":{"result":{"type":"object"}}}
tags	["education","neural-networks","autograd","visualization"]

micrograd

Interactive educational autograd engine — a TypeScript port of Andrej Karpathy's micrograd.

Features

• Value: scalar autograd node with +, ×, ÷, pow, tanh, relu, exp, and full backpropagation
• Neuron / Layer / MLP: build tiny neural networks from scratch
• Training loop: SGD with MSE, hinge, and binary cross-entropy loss functions
• Interactive walkthrough: 10-step guided tour from "what is a Value?" to "train a moon classifier"
• Canvas visualization: decision boundary grids, parameter snapshots, and training curves for the Canvas UI

Actions

Action	Description
create_session	Start a new micrograd learning session
destroy_session	Clean up a session
list_sessions	List active sessions (optionally filtered by org)
get_walkthrough_step	Get a specific walkthrough step by index
advance_walkthrough	Get the next walkthrough step in sequence
create_model	Create an MLP with a given architecture
train_model	Train a model on XOR, moon, or custom data
train_with_snapshots	Train with periodic snapshots for Canvas animation
get_decision_boundary	Generate a 2D decision boundary grid
generate_moon_dataset	Generate a moons classification dataset
get_stats	Engine statistics