FIT3206 · Conversation for artificial intelligence

Overview

In this unit you will develop applied capability in Natural Language Processing (NLP) and generative AI to build conversational systems that work in real domains with real constraints. You will progress from language fundamentals such as representations, embeddings, retrieval, and sequence modelling, to modern generative approaches (LLMs, domain adaptation, and multimodal workflows that combine text with speech/audio and, where relevant, other signals). A core emphasis is the difference between training models and delivering useful, trustworthy language systems, structured extraction, summarisation, classification, and conversation design for a specific discipline or practice. You will also learn rigorous evaluation and responsible deployment practices so your systems are robust, auditable, and appropriate for real users and stakeholders.

Offerings

The Handbook publishes no offerings for this unit.

Assessment

The Handbook publishes no assessment items for this unit yet.

Assessment details may change. Please refer to the assessment information in Moodle closer to the start of the teaching period.

Requisites

Learning outcomes

1. Diagnose and resolve complex failure modes in conversational AI (e.g., hallucinations, retrieval errors, brittleness under distribution shift, and unsafe outputs) by applying systematic error analysis and selecting justified mitigations aligned to domain risk and constraints;
2. Select, apply and justify methods and tools for building, evaluating and improving conversational AI systems, including embeddings, retrieval workflows, prompting strategies, adapters and guardrails to produce robust, auditable and context‑appropriate solutions;
3. Design and implement a domain-grounded conversational AI system by specifying user goals, domain constraints, interaction flows, and system components (e.g., retrieval augmentation, tools, and transparency mechanisms) that support reliable use in context;
4. Construct and execute evaluation workflows for conversational and generative systems using task-appropriate measures (e.g., quality, robustness, latency/cost, and auditability) and produce reproducible evidence to inform iteration and release decisions;
5. Assess and address responsible deployment requirements for conversational AI, including bias, safety, privacy-aware handling of sensitive content, and stakeholder-appropriate disclosure and monitoring practices that support accountability and trust.

Workload

No workload detail published.

Student experience, not official rules. Nothing here changes what the Handbook says.