Looking for a supervisor for your bachelor's thesis, master's thesis or PhD? If you are interested in artificial intelligence, virtual and augmented reality, computer vision or video games, get in touch. Many of the theses I supervise are part of AIR group projects with hospitals and companies, such as Smart-RehabAR or MRP-5G. You can also bring your own idea and we will shape it together.

Below are the PhD theses, master's theses, bachelor's theses and final degree projects I have supervised or am supervising, each master's and bachelor's thesis with its PDF, so you can see what we do. Titles are given in their original language.

  Propose a thesis

Open topics

• Admin platform for inclusive occupational safety training · Bachelor's thesis
  Many workers receive the safety training for their job in a language they do not master or in texts they cannot read easily. This project builds the management side of a system that delivers that training in their own language or in easy-to-read format and checks that they have understood it. The company registers its staff, assigns each person the module for their job and the version they need, tracks who has completed it and downloads a certificate recording the exact version of the content each worker saw. Since a safety specialist reviews the content, the platform keeps its versions and who approved them. By design, it stores no health or disability data.
  Technologies: web application with a REST API, relational database, PDF generation and, as an extension, delivery through WhatsApp Business.
• Benchmark for measuring AI transcription of historical manuscripts · Bachelor's thesis
  AI models can already read handwritten archival documents, but they fail in ways that are hard to spot. They get proper names and dates wrong without warning and, when the handwriting defeats them, they sometimes present as read a text that is not in the document. This project builds an open benchmark to measure it. It gathers public-domain digitised documents from different periods with their reference transcriptions, computes character and word error rates with palaeographic normalisation of old letterforms and abbreviations, detects invented text and checks whether the confidence each system reports matches its accuracy. It is then used to compare several systems, from general-purpose multimodal models to dedicated transcription tools.
  Technologies: Python, multimodal models through APIs and run locally, evaluation metrics and a small web app to browse the results.