Federated learning, NLP, RAG reliability and machine learning systems

Hallucination in Retrieval-Augmented Generation

Structured literature review examining failure modes including retrieval failure, irrelevant or conflicting evidence, context-window limitations, model overconfidence and incorrect citations. The project compares mitigation approaches such as retrieval filtering, evidence verification, confidence estimation, abstention and human review.

Core research interests

• Federated learning and privacy-preserving machine learning
• Natural language processing and transformer models
• Generative AI and Retrieval-Augmented Generation
• Reliable AI, evidence grounding and hallucination mitigation
• Efficient AI systems and machine learning performance

Assessing Infodemic in the Post-COVID-19 Risk Communication and Governance

A mixed-methods study of post-COVID-19 infodemic, risk communication and governance across Bangladesh, India and the United Kingdom.

Combining Natural Language Processing and Federated Learning for Consumer Complaint Analysis: A Case Study on Laptops

Applied federated learning to transformer-based NLP for privacy-aware consumer complaint classification, including BERT, DistilBERT and RoBERTa experiments.

Evaluating Multi-Core Performance of Machine Learning Models Across Different Computing Environments

Compared machine learning workload performance across multi-core computing environments with a focus on execution behaviour and scalability.