TorchLeet

Overview of TorchLeet

What it is

TorchLeet is a collection of 68 PyTorch problems sourced from real ML/AI interviews at companies including Google, Meta, Anthropic, OpenAI, DeepMind, and others. It provides Jupyter notebooks for each problem, an auto-grader to check solutions, and an MCP (Model Context Protocol) AI tutor that can guide learners through problems without giving away answers. The problems are organized into three tracks: Basics (24 problems), LLM Learning Path (23 problems), and Advanced (48 problems), with company tags indicating which problems are asked by specific employers.

Key features

• 68 real interview problems from companies like Google, Meta, Anthropic, OpenAI, DeepMind
• Three tracks: Basics, LLM Learning Path (build an LLM from scratch), and Advanced
• Jupyter notebooks with question files and solution files (marked _SOLN)
• AI Tutor via MCP server that provides progressive hints, company prep plans, and learning guides (torchleet-tutor, torchleet-interview-prep, torchleet-review, torchleet-explain)
• Company-tagged questions so users know exactly what each company asks
• Auto-grader to verify solutions

Use cases

• Preparing for ML/AI engineering interviews at top tech companies
• Building a complete LLM from scratch through a structured learning path
• Practicing core PyTorch fundamentals and advanced topics like attention mechanisms, alignment techniques, and GPU kernels

Getting started

bash# Clone the repogit clone https://github.com/Exorust/TorchLeet.gitcd TorchLeet# Install PyTorch (see https://pytorch.org/get-started/locally/)# Open a problem notebookjupyter notebook torch/basic/lin-regression/lin-regression.ipynbTo use the AI Tutor, connect via MCP:bash# Claude Codeclaude mcp add torchleet -- npx -y torchleet-mcp# Codexcodex mcp add torchleet -- npx -y torchleet-mcpFor Claude Desktop / Cursor / VS Code, add the MCP server configuration as described in the README.

License and pricing

This project is licensed under the MIT License and is open source (free).

Based on the project's public README and repository metadata.