# zorost/AI-Engineering-Lab: a 24-week, notebook-driven AI engineering curriculum

> A free MIT-licensed course that walks one freight case study through 43 runnable notebooks, from Python basics to Databricks. It is a syllabus with execution attached, not a prompt workshop.

**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.

- Repository: https://github.com/zorost/AI-Engineering-Lab
- Website: https://zorost.com/ai-engineering-lab
- Stars: 312 · Forks: 258
- Language: Jupyter Notebook
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/zorost-ai-engineering-lab

## The gap AI-Engineering-Lab is trying to fill

Most self-taught AI paths are a pile of disconnected tutorials: a tokenizer video here, a RAG demo there, an agent notebook that assumes a paid API key. The repository README frames the project against exactly that, calling it "not a two-hour prompt course" and describing the goal as the curriculum the Lab would hand a new engineer joining a generative AI or Databricks modernization team. The audience named in the README is specific: software developers adding AI, analysts moving toward engineering, students and career changers, and technical founders. The prerequisites are equally specific. You need a computer you can install software on, about ten hours a week, and basic computer literacy. The README states you do not need prior Python, a GPU, or a paid API key. It also names who should walk away: anyone pursuing a research career in model architecture, or looking for a weekend prompt workshop. That honesty about scope is the first thing worth weighing, because a 24-week commitment is the actual cost here, not the license.

## One freight company, 24 weeks, 43 notebooks

The organizing mechanism is a single fictional case study. You play the AI engineering team at ZoroLogistics, a freight operator, and the seeded dataset generated in Week 1 is reused through the rest of the program: Week 2 SQL practice, Week 3 training data, Week 7 retrieval corpus, Week 10 fine-tuning set, Week 16 agent tools, Week 23 feature tables. That continuity is the design decision that separates this from a course that ships 24 unrelated demos. The README justifies the domain choice plainly: freight is regulated, traceable, and full of messy operational text, and the skills transfer to aviation, manufacturing, pharma, government, and finance. The curriculum is split into seven phases: Foundations (weeks 1 to 4), LLM core (5 to 8), Model engineering (9 to 11), Harnesses and loops (12 to 13), Agents (14 to 17), Cloud AI platforms (18 to 20), and Databricks zero to hero (21 to 24). Each week follows four beats the README lays out as Study, Build, Ship, Reflect, roughly ten hours total. From Week 3 onward, the rule is that nothing is finished until it carries a metric and an error note. That single constraint is the most opinionated thing in the repository, and it is the part most likely to survive after you finish.

## Installing it and running the first week

There is no package to install. The README gives a three-command clone and dependency install, then points you at START-HERE.md and curriculum/week-01. The requirements file covers Weeks 1 to 8 only, and its own header says later weeks add extras in the first cell of each notebook.

```bash
git clone https://github.com/zorost/AI-Engineering-Lab.git
cd AI-Engineering-Lab
python -m pip install -r requirements.txt
```

That installs numpy, pandas, matplotlib, scikit-learn, jupyter, ipykernel, pyarrow, duckdb and openpyxl, with minimum versions pinned. Week 1 is where the machine gets set up and the dataset every later week reuses is generated. The README states no GPU is needed for the first eight weeks, so a laptop is enough to get started.

API keys are handled through a template rather than committed files. Copy .env.example to .env and fill in only what the current week needs. The template groups keys by phase: OPENAI_API_KEY, ANTHROPIC_API_KEY and OPENROUTER_API_KEY for Weeks 5 and later, OLLAMA_BASE_URL for a local stack with no key, then AZURE_AI_PROJECT_CONNECTION_STRING, GOOGLE_API_KEY, GOOGLE_CLOUD_PROJECT and AWS_REGION for Weeks 18 and later, and DATABRICKS_HOST, DATABRICKS_CLIENT_ID, DATABRICKS_CLIENT_SECRET and DATABRICKS_TOKEN for Weeks 21 and later.

```bash
cp .env.example .env
```

After that, open curriculum/week-01/README.md and follow it, then open the tracker in curriculum/tracking and follow the Monday row. The README says everything else is linked from the week that needs it, so you are not expected to read the whole repository before starting.

## Where the ten-hours-a-week estimate can break

The README's prerequisite list is generous: no prior Python, no GPU, no paid key. The four-beat week assumes about ten hours. Those two claims sit in tension for the absolute beginner the README also invites. Weeks 1 to 4 cover Python, data, machine learning and deep learning with an evaluation mindset, and a reader with no Python at all will spend the Study beat of Week 1 learning syntax rather than the week's topic. The README does route that reader to START-HERE.md first, which is the right mitigation, but it does not publish a separate timeline for people starting from zero. Treat the ten hours as a floor for someone who already writes code, not a promise.

The cloud phase has its own friction. Weeks 18 to 20 run one agent across Azure AI Foundry, Google Vertex AI and AWS Bedrock. The .env.example template lists connection strings, project IDs and an AWS region, which means you need accounts and credentials on three platforms to complete that phase as written. The README does not document a fallback for readers who only have one cloud account. Nothing in the repository says the phase is optional, and nothing says it is not. That is a real gap if you are evaluating this for a cohort with restricted cloud access.

The repository has no retrieved releases, so there is no versioned snapshot to pin. You get the main branch as it stands.

## The MIT license and what it does not settle

The repository is MIT licensed, and the README states there is no signup. That combination matters for two audiences. A company can fork the curriculum, strip the ZoroLogistics case study and substitute its own domain data without asking permission, provided it keeps the license notice. An individual can work through it privately with no account, no email capture and no telemetry described in the README.

MIT covers the code and notebooks. It does not cover the third-party services the later weeks call. The .env.example template implies accounts with OpenAI, Anthropic, OpenRouter, Azure, Google and AWS, and each of those has its own terms, quotas and data-handling rules. The license on this repository says nothing about what happens to data you send to those endpoints. If you are adopting the curriculum inside a regulated organization, the license question is the easy one; the vendor and data-residency question is the one that will take time.

Upgrade cost is low by construction. There is no package to bump and no server to patch. The maintenance surface is the notebooks themselves and the pinned minimum versions in requirements.txt. If a dependency releases a breaking change, the fix lands in the repository, not in your environment, unless you have forked.

## AI-Engineering-Lab compared with fast.ai and the DeepLearning.AI specializations

The closest comparison is fast.ai's Practical Deep Learning for Coders. Both are free, both are code-first, and both target people who learn by running things. The difference is in the spine. fast.ai builds a general deep learning foundation and moves top-down from working models to theory. AI-Engineering-Lab builds a single case study and moves through it in phases, ending in a governed Databricks lakehouse rather than in model architecture. The README is explicit that a research career in model architecture is not the target.

The other common path is a sequence of separate specializations from a platform vendor, one for LLMs, one for RAG, one for agents. Those give you breadth across instructors and no continuity between artifacts. Here the Week 1 dataset is still in play at Week 23. The trade-off is that you inherit one team's opinions about tooling: Claude Code, Cursor, OpenCode, DeepSeek Harness, OpenClaw and Hermes appear by name in the phase table, and if your organization has standardized elsewhere, you are translating as you go. A vendor specialization would have the same problem with a different vendor.

## Conclusion

Adopt it if you can commit roughly ten hours a week and want a single continuous case study instead of scattered tutorials; the README states no GPU and no paid API key are required for the early weeks. Skip it if you want model-architecture research or a weekend prompt course. Before starting, verify that curriculum/README.md still lists 24 weeks, that requirements.txt installs cleanly on your Python version, and that the week you plan to run has not been superseded by a later CHANGELOG.md entry.

## FAQ

### Is zorost/AI-Engineering-Lab free, and do I need to sign up?

It is MIT licensed and the README states there is no signup. You clone the repository and install requirements.txt. The later weeks do call third-party LLM and cloud services, which have their own accounts and costs.

### Do I need a GPU or a paid API key to start zorost/AI-Engineering-Lab?

The README states you do not need prior Python, a GPU, or a paid API key, and that no GPU is needed for the first eight weeks. Weeks 5 and later can use a local stack through OLLAMA_BASE_URL instead of a hosted key.

### How long does zorost/AI-Engineering-Lab take?

The program is 24 weeks, and the README describes each week as about ten hours split across Study, Build, Ship and Reflect. That estimate assumes basic computer literacy; a reader starting with no Python will likely need longer in the first phase.

## Sources

- [Issues](https://github.com/zorost/AI-Engineering-Lab/issues)
- [License: MIT](https://github.com/zorost/AI-Engineering-Lab/blob/main/LICENSE)
- [Project website](https://zorost.com/ai-engineering-lab)
- [README](https://github.com/zorost/AI-Engineering-Lab/blob/main/README.md)
- [zorost/AI-Engineering-Lab on GitHub](https://github.com/zorost/AI-Engineering-Lab)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/zorost-ai-engineering-lab