PhD Position Image Description via Language Models Blind and Visually Impaired Readers

Job description

TACIT Project

This position is based within the NWO's Perspectief project, TACIT (Inclusive Technologies for Access and Social Participation). The goal is to co‑design AI‑enabled adaptive image and schema descriptions for digitised books and collections of the National Library of the Netherlands (KB) for blind and visually impaired readers.

Around 25% of the Dutch population lives with long‑term visual, hearing, or cognitive disabilities and diversities, often facing barriers to participation, equity and inclusion. The European Accessibility Act (EAA), which mandates that EU service providers ensure digital accessibility by 2025, marks a critical step toward reducing these inequalities. However, technology that is not co‑designed with affected communities often lacks social acceptance and fails to meet real needs.

As part of the TACIT project, this PhD position will contribute to the co‑design and co‑creation of AI and immersive media technologies through inclusive Human‑Computer Interaction (HCI) and participatory design methods. TACIT aims to enhance the accessibility of public services, accelerate the adoption of the EAA, drive innovation in assistive technologies, and promote broader societal inclusion through equitable, inclusive, and human‑centered design.

The TACIT project is funded by the NWO Perspectief program.

As part of a larger research initiative, TACIT focuses on the co‑design of inclusive and accessible AI and XR technologies for people with disabilities. The TU Delft Faculty of Electrical Engineering, Mathematics & Computer Science (EEMCS) invites applications for a PhD position aimed at developing advanced AI‑driven image description systems tailored for blind and visually impaired (BVI) readers, addressing the diverse and evolving needs of users with different types and levels of visual and cognitive diversities.

Leveraging large language and multimodal models, the research will focus on developing AI technologies to generate informative, concise, and personalised descriptions that enhance accessibility across various contexts. The candidate will engage in identifying user requirements, fine‑tuning models for accessibility tasks, and designing mechanisms that automatically adapt image content to individual user profiles. The project also involves close collaboration with BVI individuals and other stakeholders through participatory design and evaluation, ensuring that the solutions are robust not only technically but also grounded in real‑world accessibility needs.

• Co‑design and co‑create AI and immersive media technologies for people with disabilities.
• Develop advanced AI‑driven image description systems tailored for blind and visually impaired readers.
• Identify user requirements and fine‑tune large language and multimodal models for accessibility tasks.
• Design mechanisms that automatically adapt image content to individual user profiles.
• Conduct participatory design, evaluation and research with BVI individuals and stakeholders to ensure technical robustness and real‑world applicability.
• Collaborate across the Web Information Systems Group, Knowledge and Intelligence Design Group and the National Library of the Netherlands.

• Master's degree (or equivalent) in Computer Science, AI or Data Science, Human‑Computer Interaction, Electrical Engineering, or a related field.
• Excellent programming and research skills.
• Fluency in Dutch is recommended but not mandatory.
• Strong interest in accessibility, assistive technology, human‑computer interaction, and co‑design research.
• Ability and willingness to work in a multidisciplinary and multi‑stakeholder team, including working with people and communities living with disabilities.
• Prior publications and experience in AI, human‑computer interaction research are assets but not a requirement.

Doctoral candidates will be offered a four‑year period of employment in principle, but in the form of two employment contracts. An initial 1.5‑year contract with an official go/no‑go progress assessment within 15 months is followed by an additional contract for the…