Module 1
LLM and Agents Foundations
What language models are, how they are built, how they learned to reason, and how they became agents.
Weeks in this module →KAUST · Computer Science · Fall 2026/2027
CS 394E Contemporary Topics in LLMs
A research-based graduate course on modern LLM systems — from foundational architectures to agentic AI in the physical world.
- Meetings
- Mondays
14:30 – 17:30 - Location
- Building 9, Classroom 3223
- Instructor
- Marco Canini
marco@kaust.edu.sa - Office hours
- By appointment
Room 4418, Level 1, Al-Khawarizmi (Bldg. 1)
About this course
The landscape of artificial intelligence (AI) has been fundamentally reshaped by the emergence of large language models (LLMs). These systems have evolved from simple text generators into agentic AI frameworks capable of reasoning, decision-making and autonomous action across high-stakes domains — from healthcare assistants to intelligent tutoring systems that play an active role in decisions affecting human welfare.
This research-based graduate course offers a comprehensive study of modern LLM systems, from foundational architectures to cutting-edge applications. We examine the technical and ethical challenges of building, deploying and evaluating LLMs across a variety of applications of interests, such as education, healthcare and scientific discovery.
The course emphasizes research and hands-on engineering in equal measure. You will develop expertise in scalable LLM design, multi-agent coordination, retrieval-augmented generation pipelines and safety-critical evaluation, alongside core competencies in critical literature analysis, independent research and technical communication. Tools and technologies in common use across the LLM ecosystem are integrated throughout.
Learning is structured around three threads that run the length of the term:
- Weekly paper discussions — at most two readings per week, each with a short summary due 12:30 the day of class, so that class time is spent arguing about the work rather than recapping it.
- Three implementation assignments — short, practical builds using industry-standard tooling.
- A semester-long research project — in teams of up to three, aiming at a publishable-quality contribution.
Textbook and materials
- Primary text: Dan Jurafsky and James H. Martin, Speech and Language Processing (3rd edition draft) — selected chapters. Freely available from the authors’ page.
- Papers: at most two per week, drawn from NeurIPS, ICML, ICLR, OSDI, NSDI, ACL, Nature and similar venues. Each is linked from the schedule.
Prerequisites and expectations
This is a discussion-driven graduate course. You are expected to arrive having read what was assigned and formed an opinion about it. Comfort with Python, basic deep learning, and reading research papers is assumed; experience with distributed systems or NLP tooling is helpful but not required.
How the grade is made up
- 50% Semester-long research project Teams of up to 3. Proposal 10%, midterm report & presentation 15%, final report & presentation 25%.
- 15% Student-led paper presentations Each student presents in 1–2 classes.
- 20% Paper reviews & class participation A ≤500-word summary per assigned reading, plus in-class discussion.
- 15% Implementation assignments Three short, hands-on assignments.
Four modules
Module 2
Reliability and Evaluation
Why models make things up, how to detect it, and how to tell whether a model or an agent actually works.
Weeks in this module →Module 3
Scalable LLM Systems
Serving and retrieval at scale, and the frameworks, protocols and infrastructure multi-agent systems run on.
Weeks in this module →Module 4
Domain Applications
Scientific discovery, healthcare, education, world models and embodied agents.
Weeks in this module →Key dates
- TBA Project teams formed · 23:59 AST Teams of up to 3. Email the instructor with your team members and a one-line topic direction.
- TBA Project proposal · 23:59 AST · 10% of the grade Problem statement, related work, planned method, evaluation plan and milestones.
- TBA Implementation assignment 1 · 23:59 AST Hands-on build. Late penalty 10% per 24 h up to 72 h, then zero.
- TBA Midterm project report · 23:59 AST · 10% of the grade Progress against the proposal, preliminary results, revised plan for the rest of the term.
- TBA Midterm project presentations · in class · 5% of the grade In-class presentation of progress. The meeting it occupies is still to be scheduled.
- TBA Implementation assignment 2 · 23:59 AST Hands-on build. Late penalty 10% per 24 h up to 72 h, then zero.
- TBA Implementation assignment 3 · 23:59 AST Hands-on build. Late penalty 10% per 24 h up to 72 h, then zero.
- Final project presentations · in class · 10% of the grade In-class presentations during week 15 (2026-12-07).
- TBA Final project report · 23:59 AST · 15% of the grade Conference-style write-up plus code and reproduction instructions.
Dates marked TBA will be announced in class and posted here.
What you will be able to do
- Conduct independent research in LLM systems, applications, or safety by formulating novel hypotheses, designing rigorous experiments, and presenting findings through technical reports and peer-review-quality presentations.
- Critically analyze and synthesize state-of-the-art LLM research by evaluating methodologies, identifying limitations, and proposing meaningful extensions through literature reviews and academic discussions.
- Design and implement scalable multi-agent LLM systems capable of coordinating specialized agents to solve complex, multi-step problems with dynamic task allocation and external tool integration.
- Evaluate and mitigate safety risks in LLM applications by detecting hallucinations, bias, and adversarial behavior, and implementing response mechanisms suitable for deployment in sensitive domains such as healthcare or education.
- Design and optimize LLM inference systems using advanced techniques such as serving architecture design, caching strategies, and distributed computing to meet real-world performance demands.
- Build and evaluate retrieval-augmented generation (RAG) systems that integrate vector databases, embedding models, and structured knowledge sources to ensure factual consistency and source attribution in knowledge-intensive tasks.
- Adapt LLMs to domain-specific applications in healthcare, education, or scientific research by addressing ethical concerns, regulatory standards, and domain expertise integration.