Virtual human AI
AI & Data
Semester programme:Minor AI for Society
Client company:De Stap
Project description
How can we design an AI-powered companion that helps low-literate individuals in the Netherlands navigate health related resources in a clear, accessible, and supportive way?
Context
The project sits at the intersection of healthcare and digital accessibility. It responds to a pressing challenge in the Netherlands, where approximately 4 million people have low literacy in Dutch, limiting their access to health related information. Existing health platforms like “De Stap” offer extensive resources, but are not designed with this target audience in mind, often being too text heavy, complex, or poorly structured for low-literate users.
Results
The key deliverable of our project is a working prototype of an AI-powered chatbot, developed in React with TypeScript, and integrated with advanced AI tools for speech-to-text (Whisper), text-to-speech (ElevenLabs), and natural language understanding (GPT-4o-mini). The chatbot is embedded into a widget that can be integrated into De Stap’s website. Its design is modular and maintainable, allowing easy extension and refinement.
Major outcomes include:
- A user guided conversation model that simplifies access to health information.
- Accessibility first design, considering low literate users with minimal prior digital skills.
- Validation through stakeholder meetings and expert feedback, confirming the alignment with real-world needs.
- Compliance with legal standards (GDPR and the EU AI Act), ensuring safe use of personal and conversational data.
- AI model selection analysis that balances cost, performance, and usability.
- Mockups and usability flows created in Figma, showcased during the AI event.
The solution achieved a TRL of approximately 5, meaning it is validated in a relevant environment but not yet tested with the actual user group. Future steps involve real world user testing, refining chatbot interactions based on feedback, and improving natural language responses and navigation accuracy.
About the project group
Our project group consists of multidisciplinary students with prior experience in software development, design, and AI. We dedicated one full semester to this project, collaborating intensively over a period of 16 weeks. Weekly team meetings, stakeholder consultations, and iterative development cycles were part of our working methodology. Each team member contributed to specific areas: AI model research, front-end and back-end development, societal and legal research, and user interface design. We followed an agile-inspired workflow, using GitHub for version control and Trello for task management. Regular feedback sessions with coaches and stakeholders guided our design iterations and kept the project aligned with user needs.
