Skip to content

AI Engineering — Study Notes (WS 25/26)

Bilingual (English + বাংলা) study notes, multiple-choice question banks, and exam preparation for the AI Engineering course at TU Braunschweig, winter semester 2025/2026.

All mathematics on this site is typeset with MathJax, and every chapter is available as a downloadable PDF.

Start here

  • Chapters — the seven core chapters, end-to-end.
  • MCQ Banks — 487 intuitive multiple-choice questions with answers.
  • Exam Prep — deep question bank, final revision, sample-exam analysis, coding practice.
  • Appendix — the original Chapter 2 source files (before merging).

The seven chapters

  1. Introduction — language models, next-word prediction, n-grams, the paradigm shift.
  2. Foundation Models — tokenization, embeddings, the Transformer, modern variants, sampling, pretraining systems, alignment, scaling laws, evaluation. (complete edition — core notes + extended topics merged)
  3. Prompt Engineering — system/user/assistant prompts, zero/few-shot, chain-of-thought, context management, structured outputs & function calling.
  4. Retrieval-Augmented Generation — the six-step pipeline, embeddings & chunking, vector stores, sparse/dense/hybrid retrieval, reranking, evaluation.
  5. Agents — the agent loop, architectures, autonomy, tool use, failure modes, reliability math.
  6. Fine-tuning — full fine-tuning, parameter-efficient fine-tuning, LoRA, QLoRA, quantization.
  7. Legal & Ethical — why AI is not "just software", the EU AI Act, GDPR, copyright, bias & fairness.

Exam profile

Property Value
Duration / points 120 minutes, 50 points, in English
Allowed aids non-programmable calculator
Fundamentals questions multiple choice, exactly one correct option
Analysis questions "Explain why…" — answer as cause → mechanism → consequence
Application questions mini-case — name the technique, give a justification, state a trade-off
Instructions use full technical terms, no abbreviations; round numbers to 2 decimal places

A quick math check — the exam loves this one: cosine similarity is

\[ \cos\theta = \frac{\mathbf{u} \cdot \mathbf{v}}{\lVert \mathbf{u}\rVert\,\lVert \mathbf{v}\rVert}. \]

If it renders as a fraction above (not raw text), MathJax is working.

বাংলা ব্যাখ্যা: এই সাইটে অধ্যায় ১–৭, প্রতিটি অধ্যায়ের MCQ ব্যাংক, আর পরীক্ষার প্রস্তুতির সব উপকরণ এক জায়গায়। উপরের সূত্রটি যদি ভগ্নাংশ আকারে দেখাও যায়, তাহলে বুঝবে MathJax ঠিকমতো কাজ করছে।

How to use this site

Use the search box (top) to jump to any concept, toggle light/dark mode with the icon in the header, and grab the Download PDF button at the top of each page for offline study.