AI-Engineering-Lab: a 24-week notebook curriculum built on one freight case study
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.
At a glance
- What is it?
- Zorost Intelligence AI Lab publishes a free, MIT-licensed AI engineering program as 43 runnable notebooks across 24 weeks, with ZoroLogistics as the single continuous case study. The structure is the strongest part; the repository's reliance on cloud consoles and paid platforms is the part to check before committing ten hours a week.
- Who is it for?
- Adopt it if you are a working developer or analyst who wants a sequenced, artifact-producing path through LLM engineering and can absorb roughly ten hours a week for 24 weeks. Skip it if you want model-architecture research, a weekend prompt workshop, or a course that runs entirely offline without cloud accounts.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 29 days ago.
- What is it written in?
- Mainly Jupyter Notebook, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem it targets: a curriculum you execute rather than read
Most AI roadmaps are reading lists. The README makes the contrast explicit: it says this is not a two-hour prompt course and describes the material as the curriculum the Lab would hand a new engineer joining a generative AI, applied machine learning, or Databricks modernization team. The stated audience is software developers adding AI, analysts moving toward engineering, students, career changers, and technical founders. The stated prerequisites are a computer you can install software on, about ten hours a week, and basic computer literacy. The README says you do not need prior Python, a GPU, or a paid API key, and that no GPU is needed for the first eight weeks. It also names who it is not for: a research career in model architecture, or a weekend prompt workshop. That is a narrower promise than most course repositories make, and it is the right one to hold the project to.
One dataset carried across 24 weeks, and why that changes the exercises
The structural decision that separates this from a folder of tutorials is the case study. You play the AI engineering team at ZoroLogistics, a fictional freight operator. According to the README, Week 1's seeded dataset becomes Week 2's SQL practice, Week 3's training data, Week 7's retrieval corpus, Week 10's fine-tuning set, Week 16's agent tools, and Week 23's feature tables. The README gives a reason for the domain: freight is regulated, traceable, and full of messy operational text, and the skills transfer to aviation, manufacturing, pharma, government, and finance. The practical consequence is that later weeks cannot be run in isolation. If you skip Week 1, the retrieval corpus in Week 7 has no source. That is a constraint, not a feature pitch, and it means the curriculum rewards sequential completion and punishes cherry-picking.
The seven phases and what each one hands you
The program is divided into seven phases. Phase 1, weeks 1 to 4, covers Python, data, machine learning, and deep learning, with what the README calls an evaluation mindset from day one. Phase 2, weeks 5 to 8, covers tokens, transformers, prompt and context engineering, retrieval, graphs, and local models. Phase 3, weeks 9 to 11, covers quantization, fine-tuning with LoRA and DPO, serving, evals, and error analysis. Phase 4, weeks 12 and 13, is about harnesses and loops: Claude Code, Cursor, OpenCode, DeepSeek Harness, and spec-driven loops. Phase 5, weeks 14 to 17, covers single agents, multi-agent systems, MCP, OpenClaw, Hermes, and agent operations. Phase 6, weeks 18 to 20, runs one agent across Azure AI Foundry, Google Vertex AI, and AWS Bedrock. Phase 7, weeks 21 to 24, is Databricks: lakehouse, Unity Catalog, PySpark, Lakeflow, AI Search, Genie, and production. The phase-4 and phase-5 entries are the most time-sensitive part of the list, because coding-agent harnesses and MCP tooling change faster than the Python and SQL foundations underneath them.
The weekly loop: study, build, ship, reflect
Each week follows four beats. Study runs Monday to Tuesday and covers the week README plus one knowledge base file. Build runs Wednesday to Thursday: run the notebooks, then change them and break one thing on purpose. Ship is Friday, when you produce one artifact, one number, and one honest note. Reflect runs Friday to Sunday: a ten question quiz, pass at eight, then tick the tracker row. From Week 3 onward, the README states that nothing is finished until it carries a metric and an error note. That requirement is the most opinionated thing in the repository. It forces every artifact to be measured, which is a habit that most self-taught portfolios lack. It also means the course is not completable by reading alone, and the tracker in curriculum/tracking is where that becomes visible week by week.
Getting it running: clone, install, open Week 1
The README gives the setup as three commands. Clone the repository, change into it, and install the dependencies:
git clone https://github.com/zorost/AI-Engineering-Lab.git cd AI-Engineering-Lab python -m pip install -r requirements.txt
From there the README directs you to START-HERE.md first if you are new to programming or to AI, then to curriculum/week-01/README.md, which it says sets up your machine and generates the dataset every later week reuses. The third step is to open the tracker in curriculum/tracking and follow the Monday row. The repository also publishes the whole curriculum as a site at zorost.github.io/AI-Engineering-Lab, where all 24 weeks are listed with objectives and filterable by phase, each linking back to its folder. Reading the site before cloning is a reasonable way to check the pacing against your own background. Note that requirements.txt is a single file for a 24-week program spanning PySpark, cloud SDKs, and fine-tuning libraries; dependency conflicts across that range are a realistic first obstacle, and the README's own advice is to consult START-HERE.md when something breaks.
Where the free claim gets complicated
The README says you do not need a paid API key. Phase 6 covers Azure AI Foundry, Google Vertex AI, and AWS Bedrock across weeks 18 to 20, and Phase 7 covers a Databricks lakehouse across weeks 21 to 24. Those are managed platforms. The README does not state which free tiers cover the notebooks, whether a credit card is required to create the accounts, or what happens to a learner whose trial expires mid-phase. Local models appear in Phase 2 and quantization in Phase 3, which suggests some weeks can run without cloud spend, but the repository does not map which notebooks are cloud-only. Treat the no-paid-API-key claim as applying to the early phases and verify the cloud weeks yourself before enrolling. This is the single largest gap between the marketing line and the material.
How it differs from a single-track alternative
A common alternative is a provider-specific course, such as the Databricks Academy learning paths or a cloud vendor's own AI engineer track. Those go deeper into one platform and assume you have already chosen it. AI-Engineering-Lab takes the opposite approach: Phase 6 deliberately runs one agent across three clouds, and Phase 7 then goes deep on Databricks. The trade-off is real. You will not reach the depth of a dedicated Databricks certification path in four weeks, and the three-cloud phase is necessarily a survey rather than a specialism. What you get instead is a comparison you can make from your own notes, plus the earlier phases on transformers, retrieval, LoRA and DPO that a platform course usually assumes rather than teaches. If your employer has already standardized on one cloud, the vendor path is the more direct route. If you are still deciding, the breadth here is the point.
Licence, maintenance and what to verify before week one
The repository is MIT licensed, with the LICENSE file at the root and the badge in the README confirming it. MIT permits commercial use, modification, and redistribution provided the copyright notice and permission notice are retained. That matters if you plan to reuse the notebooks as internal training material at a company. It does not give legal advice, and if you are repackaging the content you should read the LICENSE file itself. On maintenance: the repository is not archived, and the last push recorded is 2026-08-17. No releases were retrieved, so there is no versioned course edition to pin to; you get the main branch as it stands. The topics list includes mlops, pyspark, and vector-search, which is consistent with the phase structure but tells you nothing about how often the notebooks are revised. Before starting, check the commit history on curriculum/week-01 and the cloud weeks specifically. Those two areas carry the most risk: Week 1 because everything downstream depends on its dataset generation, and weeks 18 to 24 because platform consoles and free tiers change without notice.
Editorial conclusion
Adopt it if you are a working developer or analyst who wants a sequenced, artifact-producing path through LLM engineering and can absorb roughly ten hours a week for 24 weeks. Skip it if you want model-architecture research, a weekend prompt workshop, or a course that runs entirely offline without cloud accounts. Before starting, verify three things: that curriculum/week-01/README.md still installs cleanly on your machine, that the requirements.txt pinned versions resolve against your Python, and that the week-18 to week-20 cloud notebooks can run on free tiers without a card, because that is where the program stops being free in practice.
Community notes