An 8-week LLM curriculum that respects your laptop

What it does

This repository accompanies an eight-week paid course and provides progressive Jupyter notebooks that build from running local models to constructing an autonomous agentic AI system. Each week adds new concepts and practical projects, with setup guides and cost-monitoring advice woven throughout. It is explicitly educational courseware, not a standalone library, and expects you to follow lectures or videos for full context.

The interesting bit

The README is unusually honest about hardware limits and API pricing. It actively warns students away from Llama 3.3 because a 70B parameter model is “way too large for most home computers,” and it maps out exactly which cheap model variants to use—like gpt-4.1-nano and claude-3-haiku-20240307—to keep optional cloud API spend under a few dollars. That kind of “will it run on my laptop?” realism is rarer than it should be in AI education.

Key highlights

• Eight weekly modules in Jupyter notebooks, ending in a capstone autonomous agentic AI project
• Local-first execution via Ollama, with Google Colab links provided for GPU-intensive weeks
• Detailed cost-control guidance, including specific cheap model aliases for OpenAI and Anthropic
• Guides covering multiple providers and free alternatives (OpenRouter, Gemini) in guides/09_ai_apis_and_ollama.ipynb
• Community contribution workflow: students can submit pull requests with challenge solutions

Caveats

• The README contains an “IMPORTANT ANNOUNCEMENT - DECEMBER 2025” banner; the date is unclear and may cause confusion about which branch to use for the original videos versus the refreshed material
• This is strictly course material; without the accompanying lectures, some notebooks may lack explanatory depth
• Several setup paths require platform-specific steps (admin PowerShell on Windows, Terminal on Mac) that are described as potentially finicky

Verdict

A solid starting point if you want a structured, pragmatic path into LLM engineering that treats your hardware constraints and wallet as serious requirements. Look elsewhere if you need a reusable framework or drop-in SDK.