AI Engineering Lab: 24-Week AI Engineering Program
Find the Signal. Act with Intelligence. · Developed by Zorost Intelligence AI Lab
找准信号，凭智能行动。· 由 Zorost Intelligence AI Lab 开发。
This is the heart of AI Engineering Lab: a 24-week, week-by-week path from Python fundamentals to production lakehouse AI. Every week has a section (the phase), a category (the skill area), a use case from the running ZoroLogistics case study, runnable notebooks, and a checklist you tick off in the Excel tracker.
这里是 AI Engineering Lab 的核心：一条 24 周、逐周推进的路径，从 Python 基础一路走到生产级湖仓 AI。每一周都有一个 section（阶段）、一个 category（技能领域）、一个来自贯穿全程的 ZoroLogistics 案例的用例、可直接运行的 notebook，以及一份要在 Excel 追踪表里逐项打勾的清单。
Brand new? Read
START-HERE.mdfirst, and keepreference/GLOSSARY.mdone tab away, every term in the program is defined there in plain language.
全新上手？先读 START-HERE.md，并把 reference/GLOSSARY.md 放在旁边一个标签页——课程里的每个术语都在那里用平实的语言定义了。
How the program works
- One case study, all 24 weeks. You are the AI engineering team at ZoroLogistics, a fictional freight company. The data, the models, and the agents you build in Week 1 are reused, improved, and productionized all the way to Week 24. By graduation you have a portfolio of interconnected artifacts, not 24 disconnected demos.
- Weekly cadence (≈10 hours/week).
- Mon to Tue · Study: read the week's README and the linked knowledge-base file.
- Wed to Thu · Build: run the notebook(s), then modify and extend them.
- Fri · Use case: complete the use-case exercise: ship something concrete.
- Fri to Sun · Reflect & check off: update the Excel tracker, push to your fork.
- Evals everywhere. From Week 3 onward, every AI artifact ships with a score (a metric or an eval) and an error-analysis note. That is the core AI engineering habit this program installs.
- Local-first, cloud-later. Weeks 1 to 13 run on your laptop (no GPU needed until Week 8, and even then a small model on CPU works). Weeks 18 to 24 use free/limited tiers of the major clouds.
1）一套案例贯穿 24 周：你是虚构货运公司 ZoroLogistics 的 AI 工程团队。第 1 周构建的数据、模型和 agent 会被复用、改进并生产化，一直到第 24 周。毕业时你拥有的是一个彼此关联的作品集，而不是 24 个互不相干的演示。 2）每周节奏（约每周 10 小时）：周一到周二·学习——读本周的 README 和链接的知识库文件；周三到周四·构建——跑 notebook，然后修改并扩展它；周五·用例——完成用例练习，交付一件具体的东西；周五到周日·复盘与打勾——更新 Excel 追踪表，推送你的 fork。 3）处处有评测：从第 3 周起，每一件 AI 产物都要带一个分数（一个指标或一次评测）和一份误差分析记录。这正是本课程要植入的核心 AI 工程习惯。 4）先本地、后云端：第 1 到 13 周在笔记本上跑（第 8 周前不需要 GPU，即便到那时用 CPU 跑小模型也行）；第 18 到 24 周使用各大云的免费或有限额度。
Prerequisites
- Comfort with any programming language (Python is taught from the ground up)
- Git basics (Week 1 refreshes them)
- A laptop: 16 GB RAM recommended; Apple Silicon or NVIDIA GPU helpful from Week 8
- Optional accounts (created during their weeks): GitHub, Hugging Face, a model API key of your choice, free tiers of Azure / Google Cloud / AWS / Databricks
能用任意一门编程语言上手（Python 会从零教起）；Git 基础（第 1 周会复习）；一台笔记本（建议 16 GB 内存；从第 8 周起 Apple Silicon 或 NVIDIA GPU 会有帮助）；一些可选账号（在对应周里创建）：GitHub、Hugging Face、你自选的一个模型 API key，以及 Azure / Google Cloud / AWS / Databricks 的免费额度。
The 24 weeks at a glance
| Wk | Week title | Section | Category |
|---|---|---|---|
| 1 | Python Foundations & the AI Engineering Landscape | Foundations | Python & Environment |
| 2 | Data Engineering & SQL for AI | Foundations | Data & SQL |
| 3 | Machine Learning Fundamentals | Foundations | Classical ML |
| 4 | Deep Learning with PyTorch | Foundations | Deep Learning |
| 5 | How LLMs Work: Tokens to Transformers | LLM Core | LLM Internals |
| 6 | Prompt Engineering & the Context Window | LLM Core | Prompt & Context |
| 7 | RAG, Vector Search & Knowledge Graphs | LLM Core | Retrieval & Graphs |
| 8 | Open Models & Local Inference: GPUs, Ollama, llama.cpp | LLM Core | Local Models & GPUs |
| 9 | Quantization & Efficient Inference | Model Engineering | Quantization & Serving |
| 10 | Fine-Tuning: LoRA, SFT & DPO | Model Engineering | Fine-tuning |
| 11 | Evals & Error Analysis for AI Systems | Model Engineering | Evaluation |
| 12 | Coding-Agent Harnesses: Claude Code, Cursor, OpenCode, DSH | Harnesses & Loops | Harnesses |
| 13 | Agentic Coding Loops & Spec-Driven Development | Harnesses & Loops | Loops & Specs |
| 14 | Agent Fundamentals: The Loop, Tools & Memory | Agents | Agent Core |
| 15 | Agent Frameworks: LangGraph & the State-Graph Model | Agents | Frameworks |
| 16 | Multi-Agent Systems & MCP | Agents | Multi-Agent & Protocols |
| 17 | OpenClaw, Hermes & Agent Operations | Agents | Personal Agents & Ops |
| 18 | Azure AI Foundry | Cloud AI Platforms | Microsoft |
| 19 | Google Vertex AI & Gemini | Cloud AI Platforms | |
| 20 | AWS Bedrock & SageMaker AI | Cloud AI Platforms | AWS |
| 21 | Databricks Day Zero: Unity Catalog & the Lakehouse | Databricks Zero to Hero | Platform & Data |
| 22 | Databricks Data Engineering: PySpark, Streaming & Lakeflow | Databricks Zero to Hero | Pipelines |
| 23 | Databricks ML & GenAI: Training, Serving, Genie | Databricks Zero to Hero | ML & GenAI |
| 24 | Databricks Production: DABs, Governance & the Capstone | Databricks Zero to Hero | Production & Capstone |
Each week folder (curriculum/week-NN/) is a complete lesson:
每个周的目录（curriculum/week-NN/）都是一节完整课程：
week-NN/
├── README.md # the lesson: problem → deep concepts (tables, diagrams,
│ # worked examples, pitfalls, glossary) → notebook
│ # walkthrough → use case → sources
├── notebooks/ # 1 to 3 runnable Jupyter notebooks (Python / SQL / PySpark)
├── exercises.md # graded exercises + hints + the week's checklist
└── quiz.md # 10-question self-check with answer key (pass 8/10)The phases
Phase 1 · Foundations (Weeks 1 to 4)
Python, data engineering, classical ML, and deep learning, taught the AI engineering way: every model ships with a metric, a split, and an error analysis. Week 1 builds the ZoroLogistics synthetic-data generator you reuse forever after.
Phase 2 · LLM Core (Weeks 5 to 8)
How LLMs actually work, tokens, embeddings, attention, KV cache, then the two engineering superpowers on top: prompt & context-window engineering and retrieval (RAG + knowledge graphs). Week 8 takes you local: open models, Ollama/llama.cpp/MLX, and GPU setup from CUDA to Apple Metal.
Phase 3 · Model Engineering (Weeks 9 to 11)
Make models cheaper and yours: quantization formats and serving engines, fine-tuning with LoRA/SFT/DPO, and the discipline that decides it all, evals and error analysis (Ng: the single biggest predictor of how fast a team ships an agent).
Phase 4 · Harnesses & Loops (Weeks 12 to 13)
Become dangerous with coding agents: Claude Code, Cursor, OpenCode, and the DeepSeek Harness. Learn managed context, rules files, subagents, and the three loops (agentic coding → developer feedback → external feedback) with a spec, a verifier, and a blast-radius rule.
Phase 5 · Agents (Weeks 14 to 17)
From a hand-written ReAct loop to LangGraph state graphs, multi-agent orchestration, and MCP. Week 17 runs OpenClaw as your personal lab assistant and Hermes-class open models as agent brains, then adds the ops layer: tracing, evals, and cost.
Phase 6 · Cloud AI Platforms (Weeks 18 to 20)
The same ZoroLogistics support agent, deployed three ways: Azure AI Foundry (Microsoft), Vertex AI + AI Studio (Google), Bedrock + SageMaker (AWS). Compare capabilities, governance, and cost, and learn how to pick.
Phase 7 · Databricks Zero to Hero (Weeks 21 to 24)
The full Zorost Databricks modernization playbook: Unity Catalog, Delta Lake medallion, DBSQL, PySpark, streaming, Lakeflow Pipelines & Jobs, MLflow, feature engineering, Databricks Model Training, Model Serving + Unity AI Gateway, AI Search (Vector Search), AI functions, Genie, and Agent Bricks, then DABs, CI/CD, governance, and FinOps, ending in the ZoroLogistics Lakehouse Intelligence capstone.
Progress tracking
Download tracking/ai-engineering-lab-24-week-tracker.xlsx and open it in Excel, Google Sheets, or LibreOffice. It contains:
下载 tracking/ai-engineering-lab-24-week-tracker.xlsx，用 Excel、Google Sheets 或 LibreOffice 打开。里面包含：
- A Dashboard sheet: your name, start date, per-week completion bars, and a chart of your progress across all 24 weeks.
- One sheet per week: every checklist item with a status dropdown (☐ Not started · ▶ In progress · ✅ Done · ⏭ Skipped), automatic per-week completion percentage, and a notes column.
一个 Dashboard 工作表：你的名字、开始日期、逐周完成度条，以及覆盖全部 24 周的进度图表；每周一个工作表：每个清单项都带状态下拉（☐ 未开始 · ▶ 进行中 · ✅ 完成 · ⏭ 已跳过）、自动计算的当周完成百分比，以及一列备注。
The workbook is generated from manifest.json, see tracking/README.md.
这份工作簿由 manifest.json 生成，参见 tracking/README.md。
Certification of completion
Finish all 24 weeks (including the Week 24 capstone) and you will have:
完成全部 24 周（包括第 24 周的结业项目）之后，你将拥有：
- 43 executed notebooks (Python, SQL, PySpark)
- A portfolio: fine-tuned model, RAG agent, multi-agent system, MCP server, three cloud deployments, a governed Databricks lakehouse
- An eval harness you built yourself, the artifact that separates AI engineers from demo builders
43 个已跑通的 notebook（Python、SQL、PySpark）；一个作品集：微调过的模型、RAG agent、多 agent 系统、MCP 服务器、三次云端部署、一个受治理的 Databricks 湖仓；以及一套你自己搭的评测框架——正是这件产物把 AI 工程师和「只能做演示的人」区分开来。
Share your fork and tracker dashboard with your mentor, your team, or Zorost Intelligence.
把你的 fork 和追踪看板分享给你的导师、团队，或者 Zorost Intelligence。
© 2026 Zorost Intelligence LLC · https://zorost.com
© 2026 Zorost Intelligence LLC · https://zorost.com