PyTorch coding interview prep with 68 problems, Jupyter notebooks, and an AI tutor.
Practice 68 real interview questions from Google, Meta, and Anthropic using Jupyter and MCP.
TorchLeet is an open-source collection of 68 real ML and AI interview problems designed to help you master PyTorch from scratch. It features Jupyter notebooks, an auto-grader, and an MCP-powered AI tutor that enforces a strict no-spoilers teaching policy.
The repository structures your preparation across basics, a full LLM learning path, and advanced systems tracks. You tackle actual questions asked by Google, Meta, and Anthropic by filling in TODO blocks in Jupyter notebooks and testing your implementations against provided solutions.
It suits machine learning engineers and researchers preparing for technical interviews at top-tier labs and companies. It does not suit beginners who do not yet know Python or basic tensor operations.
Tackle 68 questions sourced directly from first-person reports at Google, Meta, and Anthropic.
Build an LLM from scratch in order, covering tokenization, positional embeddings, attention, and alignment.
Connect your AI assistant via Model Context Protocol for progressive hints and code reviews without giving away answers.
Practice hands-on by completing TODO blocks in local notebooks and comparing with solution files.
Free under its MIT licence. The only cost is your time and the compute resources required to run PyTorch locally.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
You can, but the author strongly advises against it because the goal is to deeply understand PyTorch through struggle rather than get a quick answer.
You can add the torchleet-mcp package to supported clients like Claude Code, Cursor, or VS Code using the configuration commands provided in the README.
See Zekai first in Google