From your first line of Python to shipping a real AI-driven automation —
a hands-on curriculum (self-paced or instructor-led) across Python fluency, business data science,
machine learning, AI engineering, and production.

92 runnable notebooks · 14 modules · 300+ end-of-lesson exercises · 269 in-lesson checkpoints · 100% offline

🚀 Quick start  ·  📚 Curriculum  ·  📓 How it works  ·  ▶️ Open in Colab  ·  🤝 Contributing

---

Contents

• Why this course
• What's new
• Quick start
• Choose your path
• Curriculum
• Repository layout
• Datasets
• How each notebook works
• LLM providers
• Open any notebook in Colab — one-click links to all 92 notebooks
• About
• Contributing & licence

---

⚡ Why this course

• End to end. From print("hello") to a deployed, scheduled AI automation — no gaps assumed, no steps skipped.
• Runs anywhere. One click into Google Colab, or pip install locally. Every notebook runs 100% offline — no API key, no paid service required.
• Learn by doing. 300+ exercises — including a deliberate 🐞 debug-me in each lesson — every one shipping with a worked solution and the reasoning behind it.
• Built for live teaching. Every lesson is punctuated with short ✋ Quick exercise checkpoints (~2 min each) at natural section breaks, so a class can alternate ~20 minutes of instruction with a quick hands-on pause. 269 across the course, each with a scaffolded starter and a collapsible solution — and every solution has been executed in a fresh kernel to confirm it runs.
• Modern, minimal code. Charts in 1–3 lines (pandas .plot(), seaborn, sklearn's built-in plot helpers), pipelines over boilerplate — you learn the way practitioners actually write Python today.
• Visual where it counts. Key ideas — train/test splits, k-fold cross-validation, grid search, RAG pipelines, MCP topology — come with clean diagrams embedded right in the notebooks.
• Real business problems. Churn & CLV, fraud detection, demand forecasting, customer segmentation, RAG assistants, and AI governance — not toy datasets.

---

✨ What's new

• ✋ Interactive in-lesson checkpoints. Every lesson now embeds short ~2-minute Quick exercise checkpoints at natural section breaks — 269 across the course — so you pause and do every ~20 minutes instead of reading straight through. Each ships a scaffolded starter and a collapsible solution, and every solution has been executed in a fresh kernel to confirm it runs. → How each notebook works
• 🧑‍🏫 Built for live teaching. The checkpoint rhythm — lecture ~20 min → ~2-min try → reveal — turns any lesson into an interactive class with zero prep.
• 🔌 100% offline, end to end. Every notebook — including the LLM, RAG, and agent lessons — runs with no API key via a built-in MockLLM and offline stand-ins for the heavy libraries.

---

🚀 Quick start

A taste of the style you'll be writing by Module 4 — a full, leakage-free model in a handful of lines (pipelines over boilerplate, just like real practitioners):

from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import cross_val_score

# X, y: a feature matrix and churn labels you've already loaded (see Module 4)
churn_model = make_pipeline(StandardScaler(), LogisticRegression())
auc = cross_val_score(churn_model, X, y, cv=5, scoring="roc_auc").mean()
print(f"5-fold ROC-AUC: {auc:.3f}")

No setup yet — just open a notebook and press Run:

▶︎ Google Colab — zero setup (recommended)

Click a badge to open a notebook in your browser; nothing to install.

🔑 You need a free Google account to run the notebooks. Colab gives each signed-in user a free cloud runtime, so the first time you press Run it will ask you to sign in (any Gmail account works). Without signing in you can read a notebook but not execute its cells. No Google account? Use the Local Jupyter option below instead.

• Start the full course — 00_master_onboarding.ipynb  
• Start the fast track — 00_fast_track_onboarding.ipynb  
• See it work first (5 min) — 00c_see_it_work.ipynb  

Every notebook is listed with its own Colab link in Open any notebook in Colab.

⌥ Local Jupyter

python -m venv .venv
source .venv/bin/activate         # Windows: .venv\Scripts\activate
pip install -r requirements.txt
jupyter lab

Tested with Python 3.10+. Module 0 includes an environment-check cell. The 13 appendices demo heavier libraries (PyTorch, Prophet, FAISS, LangChain, …), kept commented-out at the bottom of requirements.txt — each still runs offline via a built-in stand-in, so install them only to see the real library at work.

---

🧭 Choose your path

	🎓 Complete course	🏎️ Fast track
Scope	All 14 modules + 13 optional appendices	The essentials, condensed
Notebooks	46 lessons (+ 13 appendices)	15 (onboarding + 14 lessons)
Time	~115 hours	~15 hours
Best for	Depth — every exercise, stretch problem & capstone	A credible end-to-end pass in a few evenings
Start here	00_master_onboarding.ipynb	fast_track/

New here? 00c_see_it_work.ipynb is a 5-minute offline demo of what you'll build, and 00b_course_overview.ipynb has the full module map and an interactive time estimator.

---

📚 Curriculum

Module	Lessons	Focus
0 · Onboarding	—	Setup, orientation, 5-minute demo
1 · Foundations	1–6	Variables, control flow, lists, dicts, functions, classes
2 · Data Science	7–11	pandas, NumPy, seaborn/matplotlib, statistics, time series
3 · Real-world I/O	12–13	HTTP/APIs, SQL, data validation
4 · Machine Learning	14–16	scikit-learn, cross-validation & hyperparameter tuning, model evaluation, feature engineering
5 · Industry Applications	17–20	Churn/CLV, fraud, segmentation + recommenders, demand & maintenance
6 · AI Engineering	21–26	LLM fundamentals, prompts, RAG, agents, document processing, eval & observability
7 · Building AI POCs	27–30	Copilot setup, three POCs, RAG deep dive, vector DBs + agentic AI
8 · Agents, Tools & MCP	31–34	Agent architectures, robust tools, the Model Context Protocol, multi-agent systems
9 · NLP (Text Analytics)	35–37	Topic modeling (BERTopic, STREAM) & sentiment analysis (VADER → classical → transformers)
10 · Optional: DeepTab	38	Deep learning for tabular data — Mamba/FT-Transformer/SAINT… behind a scikit-learn API
11 · Production	39–40	Packaging notebooks into projects, scheduling
12 · CI/CD & Deployment	3 labs	Docker, Compose, GitHub Actions, registries, DNS, reverse proxies, HTTPS, Ubuntu deployment — a self-contained mini-book + 3 hands-on lab notebooks + runnable example app
13 · Capstones	41–42	Two end-to-end projects (analytics + AI assistant)
14 · Business AI	43–46	Digital transformation, architecture, AI-assisted dev, governance
15 · Optional: Django	2 labs	Wrap a model in a real web app — ORM, admin, forms, a JSON API — a self-contained mini-book + 2 hands-on lab notebooks + runnable example app

13 optional appendices (classical → deep-learning → foundation-model forecasting, PyTorch, vector stores, RAG/agent frameworks) live beside their modules — all runnable offline. Modules 9–10 are optional, reference-style tracks (text analytics + deep tabular), and Module 15 (Django) is an optional mini-book — all fully offline.

---

🗂️ Repository layout

Path	Contents
00_onboarding/ … 14_business_ai/	The complete course — 46 lessons + 13 appendices
09_nlp/, 10_deeptab/	Optional reference tracks — text analytics (NB 35–37) & deep tabular learning (NB 38)
12_cicd/	CI/CD, Docker & deployment mini-book — 14 chapters + 3 hands-on lab notebooks + a runnable example app
15_django/	Optional: Django for AI web apps — mini-book (7 chapters) + 2 hands-on lab notebooks + a runnable example app, ChurnScope
fast_track/	The fast track — 14 trimmed notebooks (~15 h)
quizzes/	11 short multiple-choice quizzes (Modules 1–9, 11 & 14)
data/	Sample CSVs (support_ops, api_log, customer_feedback) — disk copies of inline data for read_csv practice; see Datasets
slides/	Course-overview deck + lecture decks (PDF + LaTeX source)
scripts/	Helpers — validate/execute every checkpoint (test_checkpoints.py), run every notebook end-to-end, regenerate the hero banner, check NB-number references
docs/	Course-design notes (pedagogical review, module-descriptor coverage)
llm_providers.py	Unified interface to OpenAI / Anthropic / Google / Ollama (+ offline MockLLM)
previous_versions/	The legacy flat 19-notebook layout, archived

---

📊 Datasets

The course is offline-first and reproducible: almost every dataset is synthetic and generated inline from a fixed random seed, so each run produces identical data with zero downloads. One fictional world ties them together — a SaaS company running an AI customer-support operation — and its tables (customers, support tickets, API-cost logs, feedback, payments) recur from module to module, so you re-meet familiar data as the techniques get harder. Three of those synthetic tables are also dumped to data/*.csv so you can practise pd.read_csv against real files; two small real datasets (Palmer Penguins, UCI Bike Sharing) are bundled there too, so the optional 📊 try it on real data sections run offline as well; and a few lessons deliberately reach for live public APIs where that is the point.

Output convention. The quantitative teaching notebooks ship with their figures rendered, so every chart is viewable directly on GitHub and in Colab before you run a single cell. Deterministic seeds mean re-running a notebook reproduces the same figures; if you re-run and commit, keep the rendered outputs in place.

Bundled sample files

Small enough to fit on one screen and travel with the repo (see data/README.md). The first four are disk copies of data the notebooks build inline; the last two are small real datasets bundled for the optional try it on real data sections.

File	Rows	What it is	Used by
data/api_log.csv	50	LLM API request log — model, segment, quarter, tokens_in/out, latency_ms	NB 7 — Pandas fundamentals
data/support_ops.csv	60	Support-ops metrics by channel & month — tickets, automation rate, latency, satisfaction, cost	disk copy of the data NB 41 (Capstone A) builds inline
data/customer_feedback.csv	15	Labelled feedback — text, sentiment, topic	sample mirroring the inline data in NB 14 & 22
forecast.csv	28	7-day weather forecast for 4 cities, saved from the Open-Meteo API (lives at the repo root)	written by NB 12 — APIs & HTTP
data/penguins.csv	344	Real — Palmer Penguins: 3 species' bill/flipper/mass measurements, with real missing values · CC0	NB 9 — Visualization
data/bike_sharing_daily.csv	731	Real — UCI Bike Sharing: daily rentals 2011–12 with weather & calendar features · CC BY 4.0	NB 20 — Demand forecasting

Synthetic datasets, by theme

All generated inline (no downloads), grouped by the business problem they illustrate:

Theme	What's in it	Notebooks
SaaS customer churn	The course backbone — tenure, charges, support tickets, usage, contract, region, churn label (+ a revenue target)	NB 14–17, 38
LLM cost & latency logs	Support calls tagged by model & channel with tokens, cost, latency, satisfaction	NB 7–9
Support operations	Tickets across five channels (Email/Chat/Phone/Web/Social), queried in an in-memory SQLite DB	NB 13, 24, 41
Fraud / payments	One row per transaction, with planted fraud patterns (night spend, new-device takeover)	NB 18
Customer segmentation	Customers drawn from hidden archetypes for clustering + recommendations	NB 19
Demand & maintenance	Short demand series (lag & rolling features) plus predictive-maintenance signals	NB 20
Time-series forecasting	Daily product-search series with trend & seasonality (classical → Prophet → DL → foundation models)	NB 11, DS A1–A4
Customer feedback & reviews	Piles of short product reviews, support tickets and survey notes for topic models + sentiment	NB 35–37, 42
RAG document corpora	Small knowledge bases / product catalogues chunked, embedded and retrieved	NB 23, 29, 30
Invoices & documents	Synthetic messy invoices for an extraction pipeline	NB 25
Golden eval sets	Tiny labelled sets for evaluating and triaging an AI feature	NB 26
Agent / copilot data	A support copilot's lookup numbers + docs, exposed as tools / MCP resources	NB 31–34
POC & app demo data	~500 synthetic customer rows, a product catalogue, and random embedding vectors that seed the POC apps	NB 27, 28, 30
Vision & sequences (PyTorch)	Toy 8×8 "digit" images, a synthetic sequence task, and a tiny text-intent set	ML A2, A3
Business case studies	Meridian, a fictional 400-person B2B SaaS, for transformation & governance scenarios	NB 43–46

Real data & live services — the few exceptions

• scikit-learn toy sets — Iris & Wine (NB 14) and Breast Cancer (NB 15), used briefly to anchor the classic ML examples before switching to the synthetic SaaS data.
• Live public APIs (no key required) — Open-Meteo (weather — the running example) and JSONPlaceholder (a fake REST API) in NB 12; Firecrawl web scraping into a RAG-ready dataset in the I/O appendix.
• Pretrained models (models, not datasets — fetched on first use when online, each with an offline fallback) — sentence-transformers embeddings and small Hugging Face transformers in the embeddings/NLP lessons.
• MockLLM — not a dataset, but the deterministic offline model that produces the text/JSON for 16 of the AI notebooks; swap one line in llm_providers.py to call OpenAI / Anthropic / Google / Ollama instead.

Going further — real datasets to drop in

If you want to take a lesson onto real data, these fit the course's themes and mostly load in one line (cached after the first fetch):

Dataset	Load / source	Pairs with	Licence
California Housing	sklearn.datasets.fetch_california_housing()	M4 regression (14–15)	public
20 Newsgroups	sklearn.datasets.fetch_20newsgroups()	M9 topic modeling (35–36)	public
statsmodels series (CO₂, sunspots, Nile)	statsmodels.datasets.co2.load_pandas()	M2 stats & forecasting (10–11, A1–A4)	public
Telco Customer Churn	Kaggle blastchar/telco-customer-churn · 7,043 rows	M4–M5 churn (14–17), M10 (38)	IBM sample
RAG Mini-Wikipedia	load_dataset("rag-datasets/rag-mini-wikipedia") · corpus + Q/A	M6–M7 RAG (23, 29, 30)	CC BY 3.0
Twitter Financial News Sentiment	load_dataset("zeroshot/twitter-financial-news-sentiment")	M9 sentiment (37)	MIT
Adult / Census Income	sklearn.datasets.fetch_openml("adult", version=2)	M14 governance & fairness (46), M10 (38)	public
Online Retail II	UCI #352 · ~1M rows	M2 pandas-at-scale, M3 SQL (13), M5 RFM/CLV (19)	CC BY 4.0
Credit Card Fraud (ULB)	Kaggle mlg-ulb/creditcardfraud · 0.17% fraud	M5 fraud (18)	DbCL v1.0

Key-free public APIs for Module 3 (NB 12), beyond Open-Meteo: REST Countries, Frankfurter (FX rates), USGS earthquakes (GeoJSON), and Hacker News — each returns a different JSON shape to practise on.

---

📓 How each notebook works

A consistent six-part template:

🎯 Objectives + ✅ prerequisites → numbered concept sections (prose + runnable code), interleaved with ✋ quick-exercise checkpoints → 🧪 practice exercises (incl. a 🐞 debug-me) → 🧠 stretch exercises A–D → 🎁 bonus mini-project → ✅ self-assessment + 🚀 next step

Every exercise — 300+ across the course — ships with a worked solution and the reasoning behind it.

✋ In-lesson checkpoints — designed for interactive teaching

Beyond the end-of-lesson exercise bank, each lesson embeds short ✋ Quick exercise checkpoints at natural section breaks — roughly one every ~20 minutes of material. They turn a lecture into a rhythm: teach ~20 min → pause for a ~2-minute hands-on exercise → teach again.

Each checkpoint is a self-contained three-cell block:

1. Prompt — a focused, business/AI-flavoured task, solvable with only what the lesson has covered up to that point.
2. # ✍️ Your turn — a scaffolded starter cell to fill in.
3. ✅ Solution — a collapsible answer with a one-line explanation.

There are 269 of these across the course (3–4 per core lesson, 3 per fast-track lesson, 4 per CI/CD & Django lab), and every code solution has been executed in a fresh Jupyter kernel to verify it actually runs (the few file-content solutions — __init__.py, pytest, pyproject.toml — are validated by inspection). A CI workflow re-checks the structure and syntax of all 269 on every push; run the full kernel test yourself with python scripts/test_checkpoints.py --exec. They run 100% offline like everything else — no API key or network needed. Self-paced learners solve each one as they reach it; instructors use them as the built-in "pause and try" beats of a class. In the conceptual Business-AI lessons (43–46) the middle cell is a short written reflection/decision task instead of code.

🧑‍🏫 Teaching live? Lecture for ~20 minutes, then jump to the next ✋ checkpoint and give the room ~2 minutes to try it before you reveal the solution. With 3–4 per lesson, a 90-minute class gets several natural interactive breaks — no prep required.

Two more conventions you'll see throughout:

• Charts take a few lines, not a page. Plots are drawn with pandas .plot(), seaborn one-liners (hue= instead of loops, sns.heatmap(cm, annot=True) for matrices), and scikit-learn's *Display helpers — raw matplotlib appears only for final tweaks and multi-panel layouts.
• Diagrams travel with the notebook. Explanatory figures (the train/test split, 5-fold cross-validation, the GridSearchCV flow, RAG pipelines, MCP host/client/server topology, walk-forward backtesting, …) are embedded as attachments inside the .ipynb files, so they render on GitHub, in Colab, and locally with no extra image files.

---

🔌 LLM providers

Notebooks 21–26 and 42 run entirely offline with the built-in MockLLM. For real intelligence, swap one line — the unified interface in llm_providers.py supports four providers:

Provider	Class	When to use
🟢 OpenAI	OpenAILLM(model="gpt-5.4-mini")	Reliable default
🟠 Anthropic	AnthropicLLM(model="claude-haiku-4-5")	Long context, careful tone
🔵 Google	GoogleLLM(model="gemini-2.5-flash")	Cheap at scale
🟣 Ollama	OllamaLLM(model="llama3.2:3b")	Local — no internet, key, or cost

Set the matching *_API_KEY env var for hosted providers (never inline). See 06_ai_engineering/A1_llm_providers_guide.ipynb for setup and cost notes. Never commit API keys.

---

▶️ Open any notebook in Colab

Every notebook below runs in Google Colab with one click — no install, no download. Click a badge to open it. Sign in with a free Google account the first time you run a cell — Colab needs it to give you a cloud runtime.

0 · Onboarding

Notebook	Open
00_master_onboarding.ipynb
00b_course_overview.ipynb
00c_see_it_work.ipynb

1 · Foundations

Notebook	Open
01_python_basics.ipynb
02_control_structures.ipynb
03_lists_data_structures.ipynb
04_dictionaries_advanced.ipynb
05_functions_modules.ipynb
06_classes_and_oop.ipynb

2 · Data Science

Notebook	Open
07_pandas_fundamentals.ipynb
08_numpy_fundamentals.ipynb
09_matplotlib_basics.ipynb
10_statistics_basics.ipynb
11_time_series_forecasting.ipynb
A1_forecasting_classical.ipynb
A2_forecasting_prophet_libraries.ipynb
A3_forecasting_deep_learning.ipynb
A4_forecasting_foundation_models.ipynb

3 · Real-world I/O

Notebook	Open
12_apis_and_http.ipynb
13_sql_fundamentals.ipynb
A1_web_scraping_firecrawl.ipynb

4 · Machine Learning

Notebook	Open
14_sklearn_basics.ipynb
15_model_evaluation.ipynb
16_feature_engineering.ipynb
A1_pytorch_foundations.ipynb
A2_pytorch_vision_and_sequences.ipynb
A3_pytorch_fine_tuning.ipynb
A4_tabpfn_priorlab.ipynb
A5_conformal_prediction.ipynb

5 · Industry Applications

Notebook	Open
17_churn_clv_retention.ipynb
18_fraud_anomaly_detection.ipynb
19_segmentation_recommenders.ipynb
20_demand_maintenance.ipynb

6 · AI Engineering

Notebook	Open
21_llm_fundamentals.ipynb
22_ai_workflows.ipynb
23_embeddings_retrieval.ipynb
24_tools_and_agents.ipynb
25_document_processing.ipynb
26_ai_evaluation_observability.ipynb
A1_llm_providers_guide.ipynb
A2_vector_stores_survey.ipynb
A3_rag_and_agent_frameworks.ipynb

7 · Building AI POCs

Notebook	Open
27_from_setup_to_first_poc.ipynb
28_three_pocs_growing_complexity.ipynb
29_rag_pipeline_deep_dive.ipynb
30_vector_db_and_agentic_ai.ipynb

8 · Agents, Tools & MCP

Notebook	Open
31_agent_architectures.ipynb
32_designing_robust_tools.ipynb
33_model_context_protocol.ipynb
34_multi_agent_systems.ipynb

9 · NLP (Text Analytics)

Notebook	Open
35_topic_modeling_bertopic.ipynb
36_topic_modeling_stream.ipynb
37_sentiment_analysis.ipynb

10 · Optional: DeepTab

Notebook	Open
38_deeptab_tabular_deep_learning.ipynb

11 · Production

Notebook	Open
39_from_notebook_to_project.ipynb
40_scheduling_orchestration.ipynb

12 · CI/CD & Deployment (labs — mini-book module)

Notebook	Open
lab01_docker_and_compose.ipynb
lab02_ci_pipeline_github_actions.ipynb
lab03_deploy_dns_https_monitoring.ipynb

13 · Capstones

Notebook	Open
41_capstone_analytics.ipynb
42_capstone_ai_assistant.ipynb

14 · Business AI

Notebook	Open
43_digital_transformation.ipynb
44_architecture_patterns.ipynb
45_ai_assisted_software_development.ipynb
46_bpm_governance_poc_mvp.ipynb

15 · Optional: Django (labs — mini-book module)

Notebook	Open
lab01_django_in_a_notebook.ipynb
lab02_serving_a_model_with_auth.ipynb

🏎️ Fast track

Notebook	Open
00_fast_track_onboarding.ipynb
01_python_basics.ipynb
02_control_structures.ipynb
03_lists_and_dicts.ipynb
04_functions.ipynb
05_classes_basics.ipynb
06_pandas_fundamentals.ipynb
07_visualization_and_stats.ipynb
08_sklearn_basics.ipynb
09_apis_and_sql.ipynb
10_ai_workflows.ipynb
11_embeddings_and_rag.ipynb
12_tools_and_agents.ipynb
13_notebook_to_project.ipynb
14_agents_and_mcp.ipynb

🧠 Quizzes

Notebook	Open
quiz_01_foundations.ipynb
quiz_02_data_science.ipynb
quiz_03_real_world_io.ipynb
quiz_04_machine_learning.ipynb
quiz_05_industry_applications.ipynb
quiz_06_ai_engineering.ipynb
quiz_07_building_ai_pocs.ipynb
quiz_08_agents_tools_mcp.ipynb
quiz_09_nlp.ipynb
quiz_11_production.ipynb
quiz_14_business_ai.ipynb

---

👤 About

I am Prof. Dr. Christoph Weisser, Professor of Mathematics, specializing in Business Data Science at Hochschule Bielefeld (HSBI), and former Technical Lead Analytics & Artificial Intelligence at BASF. My work focuses on Artificial Intelligence, Generative AI, Business Data Science, and agentic AI systems that bridge research with real-world industrial applications.

Before joining academia, I led international AI initiatives at BASF from strategy through production deployment. Today, I combine research, teaching, open-source software development, and selected industry collaborations to advance the practical application of AI.

I completed the PhD Program in Applied Statistics & Empirical Methods (summa cum laude) at Georg-August-Universität Göttingen and studied at the University of Oxford and the University of St Andrews as a scholar of the Studienstiftung des deutschen Volkes. I regularly publish research in leading journals and international conferences and contribute to open-source AI software.

---

🤝 Contributing & licence

Spotted a bug or an unclear explanation? Open an issue or PR — contributions are welcome.

Licensed under the MIT License (see LICENSE) — use freely for learning, teaching, or anything else.

_{Happy coding 🚀}