LLM Fundamentals

Course Overview

This course covers the theoretical and practical foundations of large language models, tracing the path from RNNs to the Transformer architecture and unpacking core concepts like tokenization, embeddings, and context windows before moving into how models are actually trained and adapted: pre-training on web-scale data, fine-tuning, instruction tuning, and RLHF. From there it shifts into application-building, teaching prompting techniques from zero-shot to chain-of-thought, and walking through how to build a simple retrieval-augmented generation (RAG) application.

The throughline is that none of this stays abstract — a dedicated module has you making authenticated API calls against STACKIT AI Model Serving to run text generation and embeddings against open-source models, so you leave with a working Python application rather than just an understanding of transformer math. That matters because most LLM tutorials assume a US hyperscaler API key; this one is built around keeping your prompts, responses, and data inside European infrastructure from day one, and closes with a module on bias, hallucination, and privacy risks specific to production LLM applications.

What You’ll Learn

• Explain transformer architecture, tokenization, and embeddings well enough to reason about model behavior
• Distinguish pre-training, fine-tuning, and RLHF, and know when each is the right tool
• Write effective prompts using zero-shot, few-shot, and chain-of-thought techniques
• Build a simple RAG application backed by STACKIT AI Model Serving
• Recognize hallucination, bias, and data-privacy risks before they reach production

Modules

• Foundations of Large Language Models
• Training, Fine-Tuning, and Evaluation
• Prompt Engineering and Application Patterns
• Deploying LLMs with STACKIT AI Model Serving
• Ethics and the Future of LLMs

STACKIT documentation university.stackit.cloud View Course on STACKIT University Open the documentation