Course Syllabus

Agentic Large Language Models

Course Code: TBD · Credits: 6 EC (168 hours of study) · Institution: LIACS, Leiden University

For the full course description and teaching team, see About. This page covers prerequisites, learning outcomes, course structure, and assessment.

Prerequisites

  • Programming skills in Python and PyTorch
  • Understanding of classical machine learning (classification vs. regression, training/test sets, overfitting, decision trees/ensembles)
  • Familiarity with deep learning basics (MLPs, CNNs, RNNs, backpropagation, loss and activation functions)
  • Familiarity with reinforcement learning (sequential decision problems, agent/environment, Q-learning, DQN, PPO, SAC)
  • Familiarity with NLP basics (tokenization, word embeddings, attention, the Hugging Face ecosystem)

If you are missing some of this background, the first part of the course reviews the fundamentals, but a supplementary textbook in the relevant area is recommended.

Learning Outcomes

Upon successful completion of this course, students will be able to:

  1. Explain the architecture, training pipeline, and scaling behavior of state-of-the-art Large Language Models.
  2. Apply advanced reasoning techniques (chain-of-thought, self-reflection, test-time compute) for complex problem solving.
  3. Design and implement autonomous AI agents, including tool use, retrieval augmentation, and agentic memory architectures.
  4. Build and evaluate real-world assistants in domains such as medicine, finance, and scientific research.
  5. Analyze and construct multi-agent systems for collaborative task solving and social simulation.
  6. Critically assess the behavioral, cognitive, safety, and ethical dimensions of agentic AI systems.

Course Topics

The book — and the course — are organized into seven parts. Click a part below to see its chapters.

2
Foundations of Language Modeling with LLMs
Tokenization, embeddings, attention, the transformer architecture.
3
The Large Language Model Training Pipeline
Data, pretraining, finetuning, preference tuning, inference, benchmarks.
4
Scaling Laws and Efficient Large Language Models
Scaling laws, small/distilled models, sustainability.
5
Case Study 1: Building LLMs from Scratch
Assignment 1.
6
Reasoning Models
Chain-of-thought, self-reflection, reinforcement learning with verifiable rewards.
7
Vision, Action and World Models
Vision-language(-action) models, world models.
8
Metalearning
In-context and few-shot learning, inner/outer learning loops.
9
Interaction Styles for Agentic LLMs
Historic agent types, the five agentic interaction styles, agentic memory.
10
Case Study 2: BALROG for LLM Reasoning and Reflection
Assignment 2.
11
Search and Retrieval in Agentic LLMs
RAG, Self-RAG, agentic and multimodal retrieval.
12
How Agents Use Tools to Act in the World
Toolchains, the ReAct loop, the Model Context Protocol (MCP).
13
From Tools to Assistants: Real-World Applications
Shopping, medical, financial, scientific, and coding assistants.
14
Case Study 3: Building a Multi-Agent Diagnosis Framework
Assignment 3.
15
Studying Behavior of Agentic LLMs
Machine behavior, social/interactive competencies, Theory of Mind.
16
Role-Based Interaction in Multi-LLM Systems
Role-based collaboration, orchestration topologies and protocols.
17
Emergent Behavior in Multi Agent LLM Environments
Cooperation, norm formation, collective intelligence.
18
Case Study 4: Simulating Societies with LLM Agents
Assignment 4.
19
Synthetic Data Generation for Training Agentic LLMs
20
Mechanistic Interpretability for LLM Agents
21
Cognitive Perspectives on Agentic LLMs
—
Future Directions for Agentic LLMs

Assessment

The final course grade is the average of the grades for the four case-study assignments — there is no separate written exam:

  • Assignment 1 (Part I) — Building LLMs from Scratch
  • Assignment 2 (Parts II/III) — Reasoning and self-reflection (BALROG)
  • Assignment 3 (Part IV) — Multi-Agent Diagnosis Framework
  • Assignment 4 (Part V) — Simulating Societies with LLM Agents

Each assignment has a deadline and a retake opportunity — see the Schedule for the week-by-week timeline and exact dates.

Required Materials

  • Course textbook: Agentic Large Language Models (Van Duijn, Van der Meer, Plaat & Van Stein)
  • Lecture slides (provided online)
  • Access to LLM APIs and a Python programming environment

Academic Integrity

Students are expected to follow Leiden University’s academic integrity policies. While collaboration on concepts is encouraged, all submitted work must be original, with proper attribution for any external resources used.

Accessibility

If you require any accommodations to participate fully in this course, please contact the instructors or Leiden University’s disability services as soon as possible.

Last updated: July 2026