Syllabus — COMP429 AI Systems Engineering and Intelligent Applications

1. Course identification

COMP429 — key facts
Course code	COMP429
Course title	AI Systems Engineering and Intelligent Applications
Subject area	Computing & Information Systems · Artificial Intelligence
Level	Undergraduate · final year / advanced
Credits	15 credits, one semester
Prerequisites	COMP329 Advanced Machine Learning and Deep Learning (or equivalent)
Co-requisites	None
Delivery	2 lectures and 1 engineering studio session per week, plus supervised project work
Language of instruction	English
Primary tooling	Python 3, LangChain / LlamaIndex, vector databases, cloud AI platforms, Docker, CI/CD
Assessment	6 assignments, 24 unit quizzes, a midterm, a final exam and a major capstone project
Teaching team	Computing & Information Systems AI teaching team — contact details on the course homepage

How to use this syllabus

Sections 1–5 tell you what the course is and what it expects of you. Sections 6–7 set out the curriculum in full. Sections 8–10 explain how you are assessed and what the capstone involves. Sections 11–14 cover resources, policies and support. Dates in the schedule are indicative and confirmed on the course homepage each semester.

2. Course description

This course focuses on the engineering, integration, deployment, governance and real-world application of intelligent systems. Students learn to design, build, evaluate, deploy and maintain enterprise-scale AI solutions.

COMP429 introduces no new machine learning algorithms. Its subject is how to build with the ones you already know. By the end of COMP329 you can train a transformer, fine-tune a foundation model and implement a reinforcement learning agent. What you have not yet done is assemble those components into a complete, reliable, production-grade system that an organisation can depend on.

Topics include AI systems architecture, agentic AI, multi-agent systems, retrieval-augmented generation, AI application engineering, cloud AI services, human-AI collaboration, AI governance, and large-scale deployment. The course culminates in the development of a production-ready intelligent application.

The emphasis throughout is on engineering judgement: choosing between architectures, managing cost and latency, handling failure gracefully, and knowing when a simpler solution is the right one.

What makes this course different

Assessment is dominated by engineering work rather than examinations. Roughly half of your assessed effort is architectural — requirements, diagrams, trade-off analyses, evaluation frameworks and governance documentation — and half is implementation.

3. Curriculum position

COMP429 is the fifth course in the AI sequence. It assumes COMP329 and prepares you for COMP529. The sequence progresses from data literacy → AI literacy → machine learning → advanced AI → AI systems engineering → AI research.

COMP429 is a terminal professional course in the sequence as well as a preparation for research. Students who stop here leave with a deployed system and a portfolio; students who continue carry the systems thinking and evaluation discipline into COMP529.

4. Prerequisites and assumed knowledge

Formal prerequisite

COMP329 Advanced Machine Learning and Deep Learning, or equivalent. You should already be able to train and evaluate neural networks, work with pretrained transformers, and write well-structured Python.

You are assumed to be able to

Train, validate and evaluate neural networks independently.
Load and fine-tune pretrained transformer models.
Write well-structured, tested Python using virtual environments.
Use Git for version control, including branching and pull requests.
Read and interpret a research paper well enough to implement from it.
Reason about model performance, overfitting and evaluation metrics.

You are not assumed to have

Any prior experience with cloud platforms.
Any prior experience with Docker or CI/CD pipelines.
Any prior experience with vector databases or retrieval systems.
Any prior exposure to AI governance, risk or compliance frameworks.
Any front-end or full-stack development experience.

These are all taught from first principles within the course.

5. Learning outcomes

On successful completion of COMP429 you will be able to engineer, deploy and govern complete AI systems at enterprise scale. The outcomes are grouped into five capability areas.

AI systems design

Design end-to-end AI system architectures.
Integrate machine learning models into software systems.
Design scalable, resilient AI pipelines.

AI engineering

Build agentic AI systems with planning and tool use.
Develop retrieval-augmented generation applications.
Integrate foundation models into business applications.
Develop multimodal AI solutions.

Deployment and operations

Deploy AI systems using cloud infrastructures.
Monitor and maintain AI applications in production.
Implement MLOps and LLMOps workflows.

Governance and trust

Evaluate AI risks and governance requirements.
Design trustworthy and reliable AI systems.
Implement human-in-the-loop solutions.

Professional practice

Deliver enterprise-quality AI solutions.
Communicate technical and strategic AI decisions.
Work to professional engineering and documentation standards.

Integrated capability

Select appropriate architectures for a problem and its constraints.
Balance cost, latency, quality and reliability trade-offs.
Evaluate complete systems, not just individual models.

6. Course structure

The course is organised into eight modules containing twenty-four units, delivered as approximately 197 tutorial pages. Each unit is delivered as a sequence of tutorials: architecture diagrams, implementation examples, industry case studies, hands-on activities and a knowledge check.

Modules at a glance
Module	Focus	Units	Tutorials
1 · AI Systems Engineering Foundations	Systems thinking, AI architectures and engineering AI projects	3	24
2 · Foundation Models & LLM Engineering	Model ecosystems, prompting at scale and LLM application engineering	3	25
3 · Retrieval-Augmented & Knowledge-Based AI	Information retrieval, RAG architecture and enterprise knowledge systems	3	25
4 · Agentic AI & Multi-Agent Systems	Agent architectures, collaboration and workflow orchestration	3	25
5 · Multimodal Intelligent Systems	Cross-modal foundations, vision-language systems and conversational AI	3	24
6 · AI Deployment & Operations	Cloud AI platforms, LLMOps/MLOps and observability	3	25
7 · Trustworthy Enterprise AI	Governance, human-centred design and AI security	3	25
8 · Designing Real-World Intelligent Applications	Industry solutions, AI product development and integrated system design	3	24
Total	Eight modules · twenty-four units	24	197

7. Unit-by-unit breakdown

The table below lists every unit in teaching order with its tutorial allocation. Tutorial pages within a unit follow a consistent structure: learning objectives, motivating scenario, theory and concepts, architecture diagrams, implementation examples, industry case study, hands-on activity, knowledge check and summary.

Tutorial allocation by unit
Module	Unit	Title	Tutorials
M1	U01	From Models to Systems	8
M1	U02	AI Architectures	8
M1	U03	Engineering AI Projects	8
M2	U04	Foundation Model Ecosystems	8
M2	U05	Prompt Engineering at Scale	8
M2	U06	LLM Application Engineering	9
M3	U07	Information Retrieval Foundations	8
M3	U08	Retrieval-Augmented Generation	9
M3	U09	Knowledge Graphs and Enterprise Knowledge Systems	8
M4	U10	AI Agents	9
M4	U11	Multi-Agent Systems	8
M4	U12	Workflow Automation and Orchestration	8
M5	U13	Multimodal AI Foundations	8
M5	U14	Vision-Language Systems	8
M5	U15	Voice and Conversational AI	8
M6	U16	Cloud AI Platforms	8
M6	U17	LLMOps and MLOps	9
M6	U18	Observability and Reliability	8
M7	U19	AI Governance in Organizations	8
M7	U20	Human-Centered AI	8
M7	U21	Security and Safety of AI Systems	9
M8	U22	Industry AI Solutions	8
M8	U23	AI Product Development	8
M8	U24	Integrated AI System Design	8
Total	24 units	Eight modules	197

8. Assessment

Assessment is dominated by engineering work. Six assignments build towards the capstone, with quizzes and examinations confirming conceptual understanding.

35%

Major capstone

A complete enterprise intelligent application, designed, built, deployed and evaluated.

30%

Six assignments

Architecture, RAG, agentic AI, multimodal, AI operations and governance.

15%

Final exam

Cumulative, with an emphasis on architecture reasoning and trade-off analysis.

10%

Midterm

Modules 1–4: systems foundations, LLM engineering, RAG and agents.

10%

Unit quizzes

Twenty-four short online quizzes, one per unit, with unlimited attempts.

Assessment components and weightings
Component	Weight	Covers	Format
Capstone project	35%	Whole course	Individual or pair project: proposal, architecture, implementation, evaluation, governance review, demonstration and report
Assignment 1 — AI Solution Architecture	5%	Module 1	Requirements document, architecture diagram, justification report
Assignment 2 — Retrieval-Augmented Generation System	5%	Modules 2–3	Knowledge base, RAG pipeline, evaluation report
Assignment 3 — Agentic AI Application	5%	Module 4	Agent design, implementation, testing report
Assignment 4 — Multimodal AI Solution	5%	Module 5	Architecture, code, evaluation
Assignment 5 — AI Operations Project	5%	Module 6	Deployment pipeline, monitoring dashboard, operations report
Assignment 6 — AI Governance Assessment	5%	Module 7	Risk assessment, governance framework, mitigation plan
Midterm	10%	Modules 1–4	Written examination, 90 minutes
Unit quizzes	10%	All 24 units	Online, one quiz per unit, unlimited attempts, best score counts
Final exam	15%	Whole course	Written examination, 2 hours, architecture reasoning and trade-off analysis
Total	100%	Weightings are indicative and confirmed in the course handbook each semester.

Assignment briefs

Assignment 1 AI Solution Architecture — design an enterprise AI architecture Brief
Assignment 2 Retrieval-Augmented Generation System — build a complete RAG application Brief
Assignment 3 Agentic AI Application — build an agent with tool use and task execution Brief
Assignment 4 Multimodal AI Solution — integrate text and image processing Brief
Assignment 5 AI Operations Project — deploy and monitor an AI application Brief
Assignment 6 AI Governance Assessment — audit a complex AI system Brief

Pass requirements

Overall grade bands follow the institutional standard scheme. In addition, no capstone is accepted without a completed governance review. A system that works but cannot be shown to be safe, fair and auditable does not pass, regardless of its technical quality.

Feedback turnaround

Marks and written feedback for assignments are returned within fifteen working days of the submission deadline. Formative feedback on capstone milestones is given in the engineering studio sessions.

9. Capstone project

Design, build, deploy and evaluate a complete AI system. You choose one of four tracks and work through twelve structured development phases from proposal to final demonstration.

Choose your track

Track A

Intelligent Knowledge Assistant

Corporate assistant, policy assistant or research assistant grounded in a document corpus.

Track B

Industry AI System

Healthcare advisor, financial analysis assistant or educational tutor for a specific domain.

Track C

Multi-Agent System

Business process automation, research automation or workflow orchestration across agents.

Track D

Multimodal Application

Vision-language assistant, document understanding system or media analysis platform.

What you submit

Project proposal and stakeholder analysis
Requirements and architecture documentation
Knowledge and data design
Working AI components and engineered workflows
Evaluation framework and test results
Governance review and deployment plan
Final demonstration and report

Portfolio ready

The capstone produces a deployed system with architecture documentation and an evaluation report — exactly the artefacts employers ask for in AI engineering roles.

Project ProposalDefine the problem, the users and the value the system delivers.
Stakeholder AnalysisIdentify everyone affected and what they need from the system.
Requirements EngineeringTurn needs into testable functional and non-functional requirements.
System ArchitectureDesign the components, interfaces and data flow end to end.
Knowledge and Data DesignPlan the corpus, schemas, embeddings and retrieval strategy.
AI Component DevelopmentBuild and integrate the models, retrievers and agents you need.
Workflow EngineeringOrchestrate components into reliable, observable pipelines.
Evaluation FrameworkDefine how the whole system will be measured before you build it.
Governance ReviewAssess risk, fairness, privacy and compliance obligations.
Deployment PlanningPlan infrastructure, cost, scaling and rollback.
System TestingTest quality, robustness, safety and failure behaviour.
Final DemonstrationPresent the deployed system and defend your engineering decisions.

Scope guidance

You may propose your own capstone problem provided it fits one of the four tracks and has a genuine user, a real dataset or corpus, and measurable success criteria. Industry-partner proposals are welcome. Every proposal must be approved before you begin component development.

10. Indicative schedule

The schedule below shows a typical fourteen-week delivery. Unit order is fixed; the pacing of later modules may shift slightly depending on studio progress. Confirmed dates are published on the course homepage at the start of each semester.

Fourteen-week teaching schedule
Week	Module	Units	Tutorials	Assessment activity
1	M1	U01 From Models to Systems	8	—
2	M1	U02 AI Architectures	8	—
3	M1	U03 Engineering AI Projects	8	Assignment 1 released
4	M2	U04 Foundation Model Ecosystems · U05 Prompt Engineering at Scale	16	Capstone proposal due
5	M2	U06 LLM Application Engineering	9	Assignment 1 due · Assignment 2 released
6	M3	U07 Information Retrieval Foundations · U08 Retrieval-Augmented Generation	17	—
7	M3 · M4	U09 Knowledge Graphs · U10 AI Agents	17	Assignment 2 due · Assignment 3 released
8	M4	U11 Multi-Agent Systems · U12 Workflow Automation	16	—
9	M5	U13 Multimodal AI Foundations · U14 Vision-Language Systems	16	Midterm (Modules 1–4) · Assignment 3 due
10	M5 · M6	U15 Voice and Conversational AI · U16 Cloud AI Platforms	16	Assignment 4 released
11	M6	U17 LLMOps and MLOps · U18 Observability and Reliability	17	Assignment 4 due · Assignment 5 released
12	M7	U19 AI Governance · U20 Human-Centered AI	16	Capstone governance review due
13	M7 · M8	U21 Security and Safety · U22 Industry AI Solutions	17	Assignment 5 due · Assignment 6 released
14	M8	U23 AI Product Development · U24 Integrated AI System Design	16	Assignment 6 due · Capstone demonstration
Exam period	Revision and final examination		—	Final exam (whole course)

Weekly workload

Plan for roughly 12–14 hours per week: two lectures, one two-hour engineering studio, and the remainder in self-directed study and project work. Capstone effort tends to expand, so start early and keep your evaluation framework in view from the beginning.

11. Tools and technology

The course uses the tooling you will meet in industry: orchestration frameworks, vector stores, containerisation and major cloud AI platforms. Teaching accounts and credits are provided for the cloud platforms used. Where possible, labs are also designed to work with locally hosted open models so that cost is never a barrier to learning.

Python 3 The language for the entire course.

LangChain / LlamaIndex Orchestration frameworks for RAG and agentic applications.

Vector databases Embedding storage and similarity search at scale.

Cloud AI platforms Azure AI, AWS AI and Google Cloud AI services.

Docker & CI/CD Packaging, deployment pipelines and version control.

Observability tooling Tracing, logging, evaluation and hallucination detection.

Laboratory assets

Engineering studio sessions are supported by a set of laboratory assets you can reuse in your assignments and capstone:

Jupyter notebooks for each unit's implementation examples.
Cloud deployment labs with step-by-step infrastructure walkthroughs.
Reference RAG and agent development projects.
Architecture templates for common AI system patterns.
Evaluation frameworks for retrieval, generation and agentic quality.

12. Reading and resources

There is no single required textbook. The course draws on primary sources — papers, technical documentation and standards — because the field moves faster than textbooks can follow. Each unit provides a reading roadmap identifying what to read and why.

Core references

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — Lewis et al., 2020. The foundational RAG paper.
ReAct: Synergizing Reasoning and Acting in Language Models — Yao et al., 2022. The basis for most tool-using agent architectures.
Chain-of-Thought Prompting Elicits Reasoning in Large Language Models — Wei et al., 2022.
Lost in the Middle: How Language Models Use Long Contexts — Liu et al., 2023. Essential for chunking and retrieval design.
Model Cards for Model Reporting — Mitchell et al., 2019, and Datasheets for Datasets — Gebru et al., 2018. Documentation practice for governance work.

Standards and frameworks

NIST AI Risk Management Framework — used in Module 7 and the capstone governance review.
OECD AI Principles — for comparative governance analysis.
ISO/IEC standards on AI risk and quality management (surveyed, not reproduced).

Technical documentation

LangChain and LlamaIndex documentation, read as primary sources rather than tutorials.
Model and API documentation from the major foundation model providers.
Cloud AI platform documentation for Azure AI, AWS AI and Google Cloud AI.
Site Reliability Engineering (Google) — for the observability and reliability material in Module 6.

How to read for this course

You are not expected to memorise papers. You are expected to be able to say what problem a technique solves, what it costs, and when you would choose it over an alternative. That is the level at which assignments and examinations assess you.

13. Course policies

Use of generative AI in your own work

This is a course about building AI systems, so it would be inconsistent to ban generative AI from your workflow. You are permitted — and in some assignments encouraged — to use generative AI tools as part of your engineering practice, subject to three conditions.

Declare it. Every submission must include a short disclosure stating which tools you used and for what (for example: drafting boilerplate, generating test cases, reviewing architecture prose).
Own it. You are accountable for everything you submit. You must be able to explain, defend and modify any part of your work in a viva if asked.
Do not outsource the thinking. Architectural decisions, trade-off analyses, evaluation designs and governance judgements must be your own reasoning. Submitting an AI-generated architecture you cannot justify is a failure of the assessment, even if the diagram is correct.

Undeclared use of generative AI in reflective, analytical or governance components is treated as an academic integrity matter.

Academic integrity

All submitted work must be your own unless the brief explicitly designates it as group work. You may discuss ideas with classmates, but the code, documentation and analysis you submit must be written by you.

Reusing code from open-source libraries, documentation or tutorials is normal engineering practice and is permitted, provided you cite the source clearly and understand what the code does. Copying another student's work, or submitting work you cannot explain, is misconduct.

The capstone includes a demonstration and defence. Being unable to explain your own architecture is treated as evidence that the work is not yours.

Collaboration and group work

Assignments 1–6 are individual. The capstone may be completed individually or in a pair, at your choice. Paired capstones are assessed against the same rubric, with an additional requirement to document the division of labour and to demonstrate individual contribution during the defence.

Peer review is part of the engineering studio sessions and is formative — it does not contribute to your mark, but participation is expected.

Late submission

Assignments submitted after the deadline without an approved extension are subject to the institutional late penalty. Because assignments build on one another and feed directly into the capstone, falling behind early has disproportionate effects — talk to the teaching team as soon as you anticipate a problem rather than after the deadline.

Extensions are granted for documented medical, personal or accessibility reasons. The capstone demonstration date is fixed and cannot be extended except in exceptional circumstances.

Data, privacy and ethics in project work

Your capstone must use data you are entitled to use. Public datasets and corpora are acceptable; scraped personal data, confidential organisational documents and licensed content are not, unless you have explicit permission.

Any project involving human participants — including user testing of a deployed system — requires ethical approval before you collect data. This is covered in Module 7 and must be addressed in your governance review.

You are also responsible for the security of any API keys, credentials and personal data used during development. Never commit secrets to a repository.

Accessibility and inclusion

Reasonable adjustments are available for all assessed components. If you have an accessibility requirement — including one that affects your ability to use cloud tooling or to participate in live studio sessions — contact the teaching team early so that alternatives can be arranged well before deadlines.

All course materials are designed to meet accessibility standards. If you encounter a barrier in any tutorial, notebook or assessment, report it; it will be fixed.

Lecture and studio attendance

Lectures are recorded and made available. Engineering studio sessions are not recorded, because they are participatory: you build, you debug, you review each other's architectures. Attendance is not formally monitored but the correlation between studio attendance and capstone quality is strong and predictable.

14. Support and communication

Where to get help

Conceptual questions: Ask in the lecture or studio session, or post on the course discussion forum where other students benefit from the answer.
Technical blockers: Bring them to the engineering studio. Debugging together is faster than debugging alone, and it is what the session is for.
Assessment and extensions: Contact the teaching team directly and early. Do not wait until the deadline has passed.
Capstone supervision: Each project is assigned a supervisor at proposal approval. Expect at least three scheduled supervision meetings across the semester.

Study advice

The students who do best in COMP429 are the ones who treat the capstone as a semester-long engineering project rather than an end-of-term deliverable. Start the evaluation framework before you build the system, and keep your architecture documentation current as the design changes.

Preparing for COMP529

COMP529 is research-oriented. The systems thinking, evaluation discipline and governance awareness you develop here are the practical grounding you need before engaging critically with the research literature. Keep your capstone evaluation report — it is often the seed of a research project.

15. Document control

Syllabus version history
Version	Date	Change
1.0	Semester 1	First release. Eight modules, twenty-four units, 197 tutorials.

Living document

This syllabus is reviewed each semester. Unit content, tooling and reading lists are updated as the field moves; assessment weightings and learning outcomes are stable. Any material change during a semester is announced on the course homepage and in the lecture.

Start Unit 1 Back to course homepage