Background

Manuscripta was proposed by Professor Dean Mohamedally at UCL in collaboration with the National Autistic Society. The project originated from a practical concern raised in schools, where students were struggling to maintain focus when using iPads in the classroom. While iPads offered the accessibility advantages of a large touchscreen and intuitive tap-based navigation, they also introduced significant distractions. Students were able to switch freely between applications, access the internet and interact with content unrelated to their lessons, undermining the structured learning environments. This problem was found to be particularly acute in specialist SEN schools, where students with autism and complex learning needs require a low-stimulus, highly structured setting, making the disruptive nature of iPads an even greater barrier to effective pedagogy.

Pen and paper, the natural alternative, was not always suitable either. For some students, the physical demands of handwriting hamper engagement, and paper-based materials offer teachers no means of monitoring student progress or providing timely adaptive feedback. Schools therefore found themselves caught between two inadequate options: a digital tool that overwhelmed students with sensory input, and an analogue approach that lacked the flexibility and responsiveness of modern educational technology.

The team was given this problem and asked to design and build a solution from first principles. There were no prescribed technologies or interfaces. The scope, architecture and implementation were determined entirely by the team, guided by ongoing input from project partners and stakeholder feedback.

The Challenge

The core challenge was to develop a digital learning ecosystem that preserved the accessibility benefits of tablet-based education while eliminating the sources of distraction and sensory overload that made iPads unsuitable for SEN classrooms. Any solution would need to meet three interdependent requirements simultaneously.

First, the student-facing interface had to be calm and constrained. This meant eliminating colour, animation, notification systems and unrestricted app access — all features standard on consumer tablets but harmful in this context. Second, the system had to be genuinely useful to teachers, providing tools to create tailored learning materials quickly, monitor the classroom in real time, and respond to individual student needs without disrupting the wider group. Third, given that the schools involved handle sensitive data about vulnerable young people, the system had to operate with no cloud dependency and full compliance with data protection requirements.

A further technical constraint was that commercially available E-ink devices, while ideal for their display characteristics, lack the computational power to run generative AI models locally. This ruled out a simple single-device solution and required the team to architect a system in which AI processing was decoupled from content consumption, running on the teacher's machine and distributing results to student devices over a local network.

Our Approach

Manuscripta addresses these challenges through a two-component system. The teacher application runs on a Windows laptop with an AI-capable chipset and uses local LLM models such as Qwen and IBM Granite to generate educational materials including quizzes, worksheets and polls. The teacher can specify the topic, target age group and reading level, and the model produces content accordingly without any internet connection. Materials are then distributed to student devices over a local area network.

The student application runs on Android-based E-ink tablets, such as those manufactured by Boox and AiPaper, as well as on Kindle and reMarkable devices. The monochromatic, low-refresh display of these devices eliminates the audiovisual stimulation associated with standard tablets, and the application runs in kiosk mode to prevent students from navigating away from their assigned material. Students can annotate worksheets using a stylus, submit answers to quizzes and polls, and raise a discreet help request visible only to the teacher.

The teacher dashboard provides a live overview of all connected devices, showing each student's status and aggregating quiz and poll responses anonymously. All data remains on the local network and is never transmitted to external servers, ensuring that no personally identifiable student information is stored or processed beyond the classroom.

Manuscripta aims to reduce sensory overload for students by delivering content on monochromatic E-ink displays, which eliminates the bright colours, notifications and unrestricted app access associated with standard tablets. The system is designed to improve engagement for quieter or more anxious students by providing a discreet help request feature that alerts the teacher without drawing attention to the individual. It also supports varying reading levels and learning needs through AI-generated content that can be adjusted in complexity on demand.

For teachers, Manuscripta replaces time-consuming manual worksheet creation with an AI content pipeline that produces tailored quizzes, worksheets and polls using on-device AI models. The platform provides a real-time monitoring dashboard that shows every student's status, help requests, and device health without requiring the teacher to move around the classroom. Additionally, it enables remote classroom controls such as screen lock and session management from a single Windows application.

From a technical perspective, the project demonstrates a fully on-device AI pipeline integrated into a practical classroom tool with no cloud dependency, using local language models running on the teacher's laptop. The system is designed with a low-latency local network protocol capable of serving up to 30 simultaneous E-ink devices, and features a modular architecture where individual components, including the networking layer and AI pipeline, can be adapted for other educational or assistive technology contexts.

Acknowledgements

The team would like to thank Professor Dean Mohamedally from UCL for his supervision and guidance throughout the project. We are grateful to Marie-Louise Holmberg, Mareli Smit and other teachers from the National Autistic Society for their input on the needs of autistic learners and their feedback on the application design. We also thank Dan Cooper, a speech and language therapist at Richard Cloudesley School, for his input on accessibility requirements for SEN students. We also thank Dr Atia Rafiq, whose demonstration of e-ink workflows in a professional healthcare context informed our approach to non-Android device support.

Partners

Partners who contributed to the project include:

• IBM, who provided technical guidance on model integration and the Granite language model, used as a fallback model in the teacher application;
• Qualcomm, who supplied an Android tablet device used during development and testing;
• National Autistic Society, who contributed domain expertise on the needs of autistic learners and provided stakeholder feedback through Marie-Louise Holmberg and Mareli Smit; and
• UCL, with academic supervision provided by Professor Dean Mohamedally.