finds.dev

zorost/AI-Engineering-Lab

A free, self-paced 24-week AI engineering course: Python, machine learning, LLMs, RAG, fine-tuning, agents and MCP, Azure and Vertex and Bedrock, and Databricks. 43 runnable notebooks, one continuous case study. MIT licensed, no signup. By Zorost Intelligence AI Lab.

View on GitHubHomepage ↗

★ 316Jupyter NotebookMITupdated Aug 2026

// the take

A free, MIT-licensed 24-week curriculum that takes a beginner from Python through ML, LLM internals, RAG, fine-tuning, agents/MCP, and three cloud AI platforms plus Databricks, all tied together by one running fictional freight case study across 43 Jupyter notebooks. Built by a boutique AI consultancy (Zorost Intelligence) as the flagship of its training program. Best fit for developers or analysts willing to put in ~10 hours a week for six months, not a weekend primer.

The single running case study (ZoroLogistics) is the real design win: Week 1's generated dataset becomes Week 7's retrieval corpus, Week 10's fine-tuning set, and Week 23's feature tables, so you end with connected artifacts instead of 24 disconnected demos. Breadth is unusual for one curriculum: tokenizer/attention internals, LoRA/DPO fine-tuning, eval harnesses, coding-agent loops (Claude Code, Cursor, OpenCode), MCP, and three cloud AI platforms (Azure AI Foundry, Vertex, Bedrock) plus Databricks, all runnable rather than slides. Pedagogy is actually thought through: a glossary, 14 knowledge-base reference files, per-week quizzes with a pass threshold, and a rule from week 3 on that every artifact ships with a metric and an error-analysis note. No GPU or paid API key is needed for the first 8 weeks, so the cheapest-to-maintain part of the course is also the lowest-friction to start.

Weeks 18-24 commit you to spinning up Azure AI Foundry, Google Vertex, AWS Bedrock, and a Databricks workspace: real accounts, real credentials, real (if capped) cloud spend. 'Free tier' is doing a lot of work in that sentence compared to the local-only first half. It teaches directly against fast-moving, pre-1.0 tools in weeks 12-17: Claude Code, Cursor, OpenCode, MCP, and three vendor agent SDKs. Those APIs and CLIs change often enough that notebooks built against today's versions have a short shelf life, and there's no sign of scheduled CI running all 43 notebooks to catch drift as dependencies move. This is a single-vendor production: Zorost Intelligence wrote it, hosts it, and uses it to sell training and consulting, so the platform choices and case-study framing reflect what Zorost sells rather than a neutral survey, and there's no outside contributor base to keep it current once Zorost's own priorities move on. Six months of maintained, runnable code across this many ecosystems is a lot for what reads as a small team to sustain — worth checking the changelog and issue tracker for how current it really is before committing.

We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
316
Forks
256
Language
Jupyter Notebook
License
MIT
Last commit
17 Aug 2026

// nearby

If AI-Engineering-Lab isn't quite it.

All repos →