Product docs

Overview

These guides cover installing and using our products. Each one also has its own site, linked in its section.

• CodeReportCard: a command-line report card for code quality.
• Pebble mk1: an interactive lab that trains a small (100k-parameter) language model on your own machine, in the browser. It walks through how language models are built.
• GrindSDE: interview-prep problem sets with C++ solutions.

CodeReportCard

A zero-config code quality report card for React, JavaScript and TypeScript projects. It scores your codebase across five categories and prints an overall grade, with an optional summary roast. Requires Node.js 18 or later; MIT licensed. npm · GitHub

Quick start

# Run without installing
npx codereportcard .

# Or install globally
npm install -g codereportcard
codereportcard ./my-project

Usage

An auto-fix mode (--fix) is listed in the README as coming soon.

Scoring

Your codebase is scored across five categories: Clarity, Duplication, Dead Code, Secrets and AI Slop. Each is weighted equally at 20% by default; change the balance with weights in your config.

Configuration

Run codereportcard init to generate a config file. Configuration is read from vibecheck.config.js or vibecheck.config.mjs, .vibecheckrc.json, or a "vibecheck" key in package.json.

// vibecheck.config.js
export default {
  // Additional glob patterns to exclude
  exclude: ['**/generated/**', '**/migrations/**'],

  // Per-rule overrides
  rules: {
    'file-too-long': { limit: 400, severity: 'warning' },
    'high-complexity': { limit: 15, severity: 'error' },
  },

  // Per-category minimum score thresholds (CI fails if below)
  thresholds: { secretLeakage: 100, clarity: 80 },

  // Overall score threshold — exit code 1 if below
  failUnder: 70,

  // Category weight overrides (auto-normalized to sum to 1.0)
  weights: { clarity: 0.25, duplication: 0.15, deadCode: 0.2, secretLeakage: 0.25, aiSlop: 0.15 },
};

CI and pre-commit

Fail a build when quality drops below a threshold with --fail-under. Combine it with --json for machine-readable output:

# .github/workflows/code-quality.yml
name: Code quality
on: [push, pull_request]
jobs:
  report-card:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: 20
      - run: npx codereportcard . --fail-under 75 --json

As a pre-commit hook (for example with husky), scan only the staged files:

npx codereportcard . --staged --fail-under 70

Pebble mk1

An interactive lab that builds and trains a language model step by step, entirely on your own machine and in your browser. It runs on the CPU, so there's nothing to install and no GPU required. The app itself is titled “How LLMs Work”. Open Pebble mk1

1. Before neural nets: a simple pattern-matching chat, with no AI.
2. Inside a transformer: tokenizer, encoding and decoding, training embeddings, a XOR neural net (the feed-forward layer), softmax, and finally training a full transformer.
3. Putting it to work: a tiny Qwen-style chat model, built and trained from scratch.

In the chatbot, your first message trains the model, which takes about a minute. Sampling settings (temperature, top-p), training epochs and thread count are adjustable below the chat box. A CPU / GPU guide explains what your machine can run.

GrindSDE

The NeetCode 150 and Striver's SDE Sheet in one place: 341 problems with C++ solutions, organized by pattern and difficulty. Open GrindSDE

• NeetCode / Striver: browse each sheet by pattern and difficulty.
• Templates: battle-tested templates for common patterns.
• Mapper: map a LeetCode problem to its pattern.
• Compare: side-by-side algorithm comparisons.
• System Design and a Cheatsheet for quick review.

Support

Questions or issues? Email founder.ashutoshm@11startups.tech or book a call. For CodeReportCard bugs, open an issue on GitHub.