ASI Lab.
Advancing intelligent software engineering — where large language models, agentic and multi-agent systems, and data-driven empirical methods meet reliability, security and trust across modern software ecosystems.
Led by Prof. Jacky Keung
What we study
Agentic software intelligence — how learned, self-directed systems understand, evolve, and safeguard the software that drives real-world decision making.
LLM-Powered Software Engineering
Teaching large language models to reason about, change, and generate real code. We build LLM systems that translate, repair, synchronise, and trace requirements to code — then measure exactly how far they can be trusted.
Explore publicationsLog Analysis & Anomaly Detection
Turning noisy, industrial-scale operational logs into signals that catch failures before they spread. Few-shot meta-learning and hybrid language models that stay robust even with scarce labels and shifting workloads.
Explore publicationsAgentic & Multi-Agent Systems
Orchestrating autonomous agents that plan, act, and verify — from budgeted repository-issue resolution to test-driven agentic testing. We design protocols, roles, and retrieval so agents cooperate instead of hallucinate.
Explore publicationsEmpirical SE & Testing Strategies
Rigorous, data-driven evaluation of code, models, and test selection. We replace folklore with measurements — comparative benchmarks and effort-aware prediction under competing objectives and distribution shift.
Explore publicationsTrustworthy AI & Privacy
Making fine-tuned LLMs safe to deploy — federated learning, protection schemes, and defences against membership leakage and memorisation — so private data stays private and models stay accountable.
Explore publicationsAI for Decision Support & Security
AI that supports human judgement where stakes are high: detecting scams and fraud, hardening autonomous driving against adversarial attack, and flagging out-of-distribution inputs before they cause harm.
Explore publicationsAgentic Software Intelligence
ASI unifies the study of intelligent, autonomous software agents and the data-driven methods that measure and improve them. The lab investigates how modern learning-based systems understand code, adapt to evolving requirements and operating conditions, and stay correct, secure and trustworthy in deployment — from large language models that write and repair code, to multi-agent systems that plan and verify, to the rigorous empirical measurement that separates real progress from hype.
- LLM-Powered Software Engineering
- Log Analysis & Anomaly Detection
- Agentic & Multi-Agent Systems
- Empirical SE & Testing
- Trustworthy AI & Privacy
- AI for Decision Support & Security