Module Specification

GenAI Agent Engineering

Module Specification

GenAI Agent Engineering: Advanced LLM Workflows to Solve Real-World Problems


Pre-trained language models offer remarkable capabilities, but real-world applications often demand tailored solutions within complex workflows. This module explores advanced engineering techniques to customise and integrate large language models into sophisticated systems.

You will begin by integrating APIs and managing dynamic conversation states using both commercial and open-source models. The module examines methods for maintaining context in multi-turn interactions, including short-term and long-term memory types. You will learn task decomposition, iterative prompting, and recursive query generation to guide models in delivering accurate, contextually relevant responses. You will develop agents using frameworks like LangChain and LlamaIndex, integrating custom tools and implementing human-in-the-middle workflows. Additionally, you will delve into multi-modal integration, working with models that process text, images, and audio.

Mastering these techniques equips you to build robust, contextually aware AI systems for complex, real-world applications. You will address critical concerns like data privacy and bias, implement secure database integrations, and perform advanced fine-tuning methods. This expertise positions you at the forefront of AI development, ready to engineer sophisticated solutions that meet the evolving demands of modern industries.


Mode(s) of Study Code CATS Credits ECTS Credits Framework HECoS code
Full-time Blended Learning
Part-time Blended Learning		 LL71 30 15 FHEQ - L7 artificial intelligence (100359)

Prerequisites and Co-requisites

None

Learning Outcomes

Code Attributes developed Outcomes
LO1 Knowledge and Understanding Demonstrate comprehensive knowledge of advanced LLM customisation techniques and integration in multi-step interaction workflows.
LO2 Intellectual Skills Analyse ethical implications and data privacy challenges in human-LLM collaboration and workflow integration.
LO3 Intellectual Skills Critically evaluate methodologies for managing conversation states and task decomposition in complex multi-turn LLM interactions.
LO4 Technical/Practical Skills Develop custom LLM agents utilising frameworks like LangChain to autonomously manage tasks with human oversight inclusions.
LO5 Technical/Practical Skills Implement secure database integrations with LLMs, ensuring data security and privacy through advanced anonymisation techniques.

Content Structure

Week Topic
Week 1 Introductory lecture
Introduces the module, outlining its relevance to the field and connections to other topics. It provides an overview of the content structure, key references, and assessment details.
Week 2 LLM Integration Techniques
Master the integration of Large Language Models into your applications to unlock the potential of AI in processing and generating human language. Understand why seamless integration is crucial for developing intelligent systems that interact naturally with users. By effectively connecting LLMs with your software, you lay the groundwork for advanced functionalities, enabling your applications to leverage state-of-the-art language capabilities and setting the stage for more complex AI workflows.
Week 3 Managing Conversation Context
Learn to manage conversation context effectively to maintain natural and coherent interactions with users over multiple turns. Discover why handling conversation state is vital for creating applications that remember prior exchanges and provide contextually relevant responses. By mastering techniques for tracking and utilising conversation history, you enhance user experience and enable your systems to engage in more meaningful, human-like dialogues.
Week 4 Advanced Prompt Engineering
Explore advanced prompting techniques to guide LLMs in generating accurate and relevant responses. Understand why crafting effective prompts is essential for eliciting desired behaviours from language models. By mastering task decomposition, iterative prompting, and recursive query generation, you can better control LLM outputs, ensuring your applications provide valuable and context-appropriate information to users.
Week 5 Developing LLM Agents
Delve into building intelligent agents that leverage LLMs for autonomous task handling. Learn why agent development is key to creating systems that can perform complex tasks without constant human guidance. By integrating frameworks like LangChain and LlamaIndex, you enable your applications to use custom tools, manage sub-tasks, and intelligently determine when to involve human intervention, enhancing efficiency and functionality.
Week 6 Incorporating Human Oversight
Understand the importance of integrating human oversight into LLM-driven workflows to ensure accuracy and reliability. Explore why incorporating strategies for human validation and feedback enhances your applications' performance. By designing Human-in-the-Middle workflows, you balance automation with human expertise, addressing critical concerns like data privacy and bias while improving the overall quality of your AI systems.
Week 7 Multi-Modal LLM Integration
Expand your applications' capabilities by integrating multi-modal LLMs that process text, images, and audio. Discover why leveraging models like CLIP and GPT-4 enriches user interactions and enables more comprehensive data analysis. By incorporating multi-modal inputs, you create versatile AI systems that can interpret and respond to a variety of information sources, meeting diverse user needs.
Week 8 Semantic Retrieval and Embeddings
Master semantic retrieval techniques to enable your applications to understand and process large volumes of data effectively. Learn why using embeddings and vector storage is crucial for capturing semantic similarity and enhancing information retrieval. By comparing embedding models and exploring vector stores like FAISS, Chroma, and Pinecone, you equip your systems to provide accurate and relevant responses based on extensive datasets.
Week 9 Secure Data Integration
Prioritise data security and privacy when integrating LLMs with databases. Understand why managing data flow securely is essential for protecting sensitive information. By employing techniques such as Reversible Anonymisation, Data Masking, and Tokenisation, you ensure that your applications handle data responsibly, maintaining user trust and complying with legal and ethical standards.
Week 10 Fine-Tuning and Deployment
Develop expertise in fine-tuning LLMs to customise their behaviour for specific applications. Learn why fine-tuning methods like instruction tuning, SFT, LoRa, and PEFT are vital for optimising model performance. By mastering frameworks like Hugging Face and NVIDIA's NeMo, and understanding distributed fine-tuning techniques, you prepare to scale and deploy AI solutions that are tailored to your needs.

Student Workload

The methods of teaching and learning for this module are based on the School's Technical 30 teaching system, consisting of the following activities.

Activity Total hours
Introductory lecture 1.50
Concept learning (knowledge graph) 36.00
AI formative assessment 18.00
Workshop/Lab Sessions 27.00
Independent reading, exploration and practice 153.50
Summative assessment 64.00
Total: 300.00

Teaching and Learning Methods

Activity Description
Introductory lecture

This is the first weekly session, dedicated to providing a comprehensive introduction to the module. The module leader will present an overview of the subject, elucidating its importance within various digital engineering professions and its interrelation with other modules. Students will need no preparation ahead of attending this session.

The module leader will provide a structured breakdown of the content to be covered in the subsequent 9 sessions. Students will also receive an outline of the essential reference materials, alongside suggestions for supplementary reading. The format and criteria for the summative assessment will be delineated, followed by a dedicated period for questions and answers.

A recording of the session will be available to facilitate async engagement for any other student who missed the class, also offering an opportunity to review the content again.
Concept learning (knowledge graph)

Our institution's approach to teaching is primarily based on flipped learning. Ahead of each weekly session (Workshop/Lab), students will be required to study the essential concepts that are used in the coming session so they are familiar with the theories and ideas related to that session. The study material will be in the form of written content, illustrations, pre-recorded lectures and tutorials, and other forms of content provided through the AGS.

This content is self-navigated by the students, accommodating different learning styles and schedules, allowing students to watch or listen to them at their own pace and review them as needed.
AI formative assessment

Once each concept of the theory is studied, students will be prompted to engage in formative assessment with instant AI feedback. They include multiple-choice questions, socratic questions and answers, written questions, role-play and other AI-assisted practice scenarios.

The purpose of this automated formative assessment is to provide students with immediate feedback on their understanding of module material and highlight any areas that need support or further study. They are also used to track student progress, boost motivation and promote accountability.
Workshop/Lab Sessions

Those studying in the blended learning mode will attend these 9 weekly classes (in person or remotely) during weeks 2 to 10. These sessions will complement the theory already studied during the preceding week (in our flipped-learning model), with discussions, analysis, practice or experience . They will be interactive and participatory, rather than one-way lectures. There will also be an opportunity for Q&A in every session. Depending on the nature of the content, challenges and learning activities will be pre-designed to apply flipped learning. They may include hands-on project work, group discussions or debates, roleplay, simulation, case studies, presentations, and other learning activities and opportunities. These workshops present an opportunity to apply critical thinking and problem-solving skills. They also encourage collaboration and foster a sense of community among students.
Independent reading, exploration and practice

This activity challenges students to engage with the reference material and independently explore and analyse academic literature related to the course topic. Students are expected to select relevant sources, practice critical reading skills, and where applicable technical skills, and synthesise information from multiple references. This is an opportunity to enhance research abilities, critical thinking, and self-directed learning skills while broadening and deepening subject knowledge.
Summative assessment

Summative assessments are used to evaluate student learning at the end of a module. These assessments can take many forms, including exams, papers, or presentations. Instructors can use summative assessments to measure whether students have achieved the learning outcomes for the module and provide them with a sense of their overall progress. Summative assessments can also be used to evaluate the effectiveness of the teaching methods used in the module.

Assessment Patterns

Weighting Format Outcomes assessed
60% Technical Analysis and Solution Assessment
This assessment requires students to develop a solution to a complex problem within a simulated domain, followed by a detailed analysis and reflection on their design and its theoretical underpinnings. The aim is to assess students' abilities to design practical solutions, critically analyse their work, and articulate their understanding of the technical and theoretical aspects of the module.		 K LO1
I LO2
I LO3
T LO4
T LO5
40% Invigilated Exam
This is a time-limited and closed-book exam with a mix of multiple-choice and analytical written questions that students undertake during the summative assessment period as scheduled under the School’s remote invigilation conditions to ensure quality and academic integrity.

The exam enables the students to demonstrate their successful attainment of the module learning outcomes, primarily related to knowledge and understanding, and secondarily related to Professional/Transferable Skills.

The analytical written questions will consist of problem questions representing issues and dilemmas students are likely to encounter in professional life and students have to synthesise and apply what they have learnt on the module in order to produce sound and reasoned judgements with respect to the problem.

To enable the students to practice and prepare, various formative assessment activies, including quizzes and a AI-augmented assignments and mock exams are built into the module. Additionally, throughout the course, students will regularly receive feedback on their knowledge and assignments from AI as well as peers and staff to indicate how to improve future work and how to give constructive feedback to others.		 K LO1
I LO2
I LO3

References/Indicative Reading List

Importance ISBN Description
Core Textbook 9781040134306 Atkinson-Abutridy, J. Large Language Models: Concepts, Techniques and Applications. CRC Press, 2024
Core Textbook 9781501520600 Campesato, Oswald. Large Language Models: An Introduction. Mercury Learning and Information, 2024
Supplementary Reading 9780138199197 Ozdemir, Sinan. Quick start guide to large language models: strategies and best practices for using ChatGPT and other LLMs. Addison-Wesley Professional, 2023
Supplementary Reading 9783031656460 Kamath, Uday, Kevin Keenan, Garrett Somers, and Sarah Sorenson. Large Language Models: A Deep Dive. Springer, 2024
Supplementary Reading 9781800200883 Babcock, Joseph and Bali, Raghav. Generative AI with Python and TensorFlow 2: Create images, text, and music with VAEs, GANs, LSTMs, GPT models and more: Harness the power of generative models to create images, text, and music. Packt Publishing, 2021
Supplementary Reading 9789027259158 Nolan, Brian and Periñán-Pascual, Carlos. Language Processing and Grammars: The role of functionally oriented computational models. John Benjamins Publishing Company, 2014
Supplementary Reading 9781803247335 Rothman, Denis. Transformers for Natural Language Processing: Build, train, and fine-tune deep neural network architectures for NLP with Python, PyTorch, TensorFlow, BERT, and GPT-3. Packt Publishing, 2022
Supplementary Reading 9798868800177 Amaratunga, T. Understanding Large Language Models: Learning Their Underlying Concepts and Technologies. Apress, 2023
Supplementary Reading 9789355519658 Khandare, SS. Mastering Large Language Models. BPB Publications, 2024.
Supplementary Reading 9781040052174 Thakurm K. Barker, H. Khan Pathan, A. Artificial Intelligence and Large Language Models: An Introduction to the Technological Future. Chapman and Hall/CRC, 2024

Programmes Linked to This Module

Programme Term Type
1 MSc AI and Machine Learning 2 Core
2 MSc Software Technical Leadership 2 Optional

Module Contacts

Role Description Name Email
Module Leader Oversees the overall scope, content, timeliness, and integrity of the module.

Module Approval

Stage Version Date of approval Authority Chair Revalidation
Compliance 1.0 Academic Board Dr Paresh Kathrani
Pre-Teaching 1.0 Director of Education Dr Paresh Kathrani
Note: The information detailed within this record is accurate at the time of publishing and may be subject to change.
Module Spec: GenAI Agent Engineering: Advanced LLM Workflows to Solve Real-World Problems (LL71)