Engineering LLM-Based Agents and Applications (ELLA-H) - COM00071H

«Back to module search

Module summary

The ELLA module focuses on the practical development of intelligent agents and applications powered by Large Language Models (LLMs). Students will explore how LLMs can be orchestrated and integrated with external tools, APIs, and structured memory systems to enable decision-making, problem-solving, and autonomous interaction across complex software engineering tasks. A variety of application areas will be considered, for example including the use of LLM-based technologies to develop assistive systems for medical diagnosis, or to streamline resource allocation, hazard analysis and decision-making in environmental management scenarios such as waste reduction. Core agent architectures, and retrieval-augmented agents will be examined through applied coding labs and project-based learning, and students will gain hands-on experience with agentic frameworks and prompt-augmentation techniques to build agents for a variety of purposes: searching the web, querying databases, controlling devices, composing workflows etc. Emphasis will be placed on the integration of LLMs in software engineering processes and practices, and on how to document this usage. Taking a systems-centric view, students will learn the importance of understanding and bounding the behaviour of LLM-based agents in the wider system context: how to evaluate and monitor agents’ behaviours, and how to use guardrails to frame their autonomy where necessary. Ethical and safety considerations, including agent alignment, hallucination control, and failure recovery, will also be addressed to ensure responsible deployment in real-world contexts.

Related modules

Pre-requisite modules

• Deep Learning (COM00049H)

Elective Pre-Requisites

These pre-requisites only apply to students taking this module as an elective.

Deep learning (COM00049H) or equivalent knowledge.

Module will run

Occurrence	Teaching period
A	Semester 2 2026-27

Module aims

The ELLA module aims to give students practical experience of the development of intelligent agents and applications powered by LLMs. The focus will be on addressing significant real-world challenges, in spheres such as Health and the Environment.

In addition to providing hands-on experience with up-to-date tools and techniques, the module will emphasise the place of agentic AI systems in robust software engineering methodologies, using a systems thinking approach. The focus is on agentic frameworks which allow for explicit management of agent state, persistence, failure recovery and integration with the wider systems. In particular, we will consider strategies to evaluate and monitor agents’ behaviours, and to bound their autonomy where the application requires it - for example, in systems where the output could contribute to an unsafe system state or to the propagation of information which could lead to harm. To this end, specific hazards - such as hallucination - will be explained and strategies to limit their effects will be introduced.

Module learning outcomes

• Explain and compare key agent architectures and design patterns that combine LLMs with planning, memory, retrieval, and (software engineering) tool use.

• Develop and deploy intelligent agents.

• Design autonomous agent workflows that combine LLM reasoning with structured retrieval and action execution, including multi-step planning.

• Implement real-world agent applications, such as task assistants, multi-modal planners, or decision-support systems, integrating external APIs and services.
• Evaluate agent behaviour using qualitative and quantitative metrics, and analyse limitations including tool reliability, latency, and system robustness.

• Identify and address safety, alignment, and ethical risks in agent design, including hallucinations, unintended behaviour, and over-reliance on automation.

• Explain the importance of systems thinking in evaluating the performance of LLMs in practice, in particular with respect to dependable systems.

Indicative assessment

Task	% of module mark
Essay/coursework	100.0

Special assessment rules

None

Indicative reassessment

Task	% of module mark
Essay/coursework	100.0

Module feedback

Feedback is provided through work in practical sessions, and after the final assessment as per normal University guidelines.

Indicative reading

TBD