Project Ideas

Undergraduate and Masters project ideas in Agentic AI, Privacy-Preserving Machine Learning, Differential Privacy, Model Inversion & Poisoning Attacks, and LLM Security / Safety / Auditability / Trustworthiness

If you are interested in any of the projects below, or if you have your own idea that aligns with my research interests, please get in touch. The best way to reach me is via email SunderAli.Khowaja@dcu.ie.

Important
These ideas are starting points for discussion. A good project is usually refined over several meetings.

COMSCI 3rd Year Projects

  1. Basic Membership Inference Attack on a Differentially Private Classifier

  2. Prompt-Injection Demo and Defence for a Local LLM

  3. Federated Learning with Privacy Accounting

  4. Lightweight Jailbreak Classifier

COMSCI 4th Year Projects

  1. Privacy–Utility Trade-off of Differential Privacy in Parameter-Efficient Fine-Tuning

  2. Model Inversion Attacks and Defences on Image Classifiers

  3. Data Poisoning in Federated Learning and Robust Aggregation

  4. Systematic Evaluation of Prompt Injection and Jailbreaks on Open-Source LLMs

  5. Audit Logging and Basic Safety Monitoring for an LLM Agent

COMBUS 4th Year Projects

  1. Privacy-Preserving Customer Analytics Prototype

  2. Safe LLM Assistant for Enterprise Document Q&A

  3. Lightweight Multi-Agent Workflow with Safety Guards for a Business Process

  4. Poisoning-Resilient Federated Learning Demo for Cross-Organisation Analytics

MCM (MSc in Computing) Thesis / Practicum Projects

  1. Differential Privacy in Parameter-Efficient Fine-Tuning of Mid-Sized LLMs

  2. Defending against Model Inversion and Membership Inference in Federated Learning

  3. Automated Red-Teaming and Adaptive Defences for Prompt Injection and Jailbreaks

  4. Agentic LLM Systems with Verifiable Safety Constraints and Auditability

  5. Trustworthiness Metrics and Interventions for LLM Agents

Looking forward to working with motivated students!