Projects

Multilingual Knowledge Alignment using LLMs and Knowledge Graphs

Horizon Europe

Developed a multilingual reasoning pipeline integrating transformer-based large language models with domain-specific knowledge graphs to detect semantic drift across languages and improve contextual consistency in machine translation systems.

Technologies: PyTorch, Hugging Face Transformers, Neo4j

Graph-RAG for Ethics and IPR in Virtual Worlds

Designed a retrieval-augmented generation framework combining structured knowledge graphs with generative language models to answer ethical, intellectual-property, and policy questions across multilingual datasets. Hybrid symbolic-neural reasoning improved retrieval precision by 34%.

Focus: Graph-RAG, ethical governance, intellectual property, multilingual retrieval

Language Quality Estimation using Generative Models

Built a quality-estimation model for translated text using fine-tuned T5 and GPT architectures trained on synthetic multilingual data. Integrated uncertainty estimation for confidence scoring, improving automatic localisation-review accuracy by 28% in simulation.

Technologies: T5, GPT architectures, multilingual synthetic data, uncertainty estimation

Digital Twin Simulation Pipeline

Huawei Research Ireland

Modelled AI-driven digital-twin environments for real-time cloud networks using distributed training and synchronisation across large datasets. Implemented model versioning, monitoring, and continuous-deployment workflows.

Technologies: Azure Machine Learning, MLflow, distributed training, continuous deployment

Predictive Analytics for Automotive Retail

Zerve AI

Developed and deployed machine-learning models to predict vehicle-sales velocity from market dynamics and consumer behaviour. Supported production scalability through automated data pipelines, CI/CD workflows, and explainable dashboards for non-technical stakeholders.

Technologies: PyTorch, Azure Data Factory, CI/CD, explainable analytics