Requirements

Partner Introduction

SmolPC 2.0 sits within both a direct partner context and a wider technology ecosystem. The primary external partner context for this project is MotionInput Games, while the organisations below remain important because they shaped the deployment environment, hardware targets, and runtime technologies used by the final system.

MotionInput Games

MotionInput Games is an official AI spinout of UCL Computer Science focused on accessible, touchless gaming and equitable computing. In the context of SmolPC 2.0, it provides the clearest direct external partner and publishing context through our supervisor, Prof Dean Mohamedally, who is listed by MotionInput Games as CEO & CTO. We therefore treat MotionInput Games as the project's primary external partner, while Intel, Microsoft, and IBM are better understood as technology and ecosystem-context contributors rather than day-to-day implementation partners.

Intel

Intel is relevant to SmolPC 2.0 primarily because OpenVINO and Intel hardware support informed part of the local runtime strategy. The project uses OpenVINO where it is the best supported option on the target Windows device, but Intel was not directly embedded in day-to-day development of the final implementation.

Microsoft

Microsoft is relevant mainly because SmolPC 2.0 is a Windows-first desktop application and part of the runtime path relies on Windows-compatible tooling such as DirectML and ONNX Runtime packages. Microsoft did not have direct hands-on involvement in the final implementation.

IBM

IBM forms part of the wider partner context around the project, but it was not a direct technical contributor to the final SmolPC 2.0 implementation. Its role in this report is therefore contextual rather than implementation-specific.

Project Background

Artificial intelligence is transforming education, yet access to AI-powered learning tools remains deeply unequal. Most mainstream systems still depend on cloud accounts, reliable internet connectivity, and platform lock-in, which makes them difficult to deploy in privacy-sensitive or under-resourced educational settings. According to UNESCO, the majority of primary schools and a significant proportion of secondary schools worldwide lack internet connectivity. [1] At the same time, teachers are under increasing workload pressure [2], and schools risk falling further behind in the emerging AI divide. [3]

The project began from an earlier SmolPC LibreOffice prototype that demonstrated local AI assistance was possible, but also exposed important limitations. It centred on one application, relied on a more cumbersome local inference setup, and did not translate cleanly into a broader classroom platform. As the project scope expanded to include a dedicated coding workflow together with GIMP and Blender support, it became clear that simply extending the old architecture would create repeated deployment, security, and maintainability problems.

Those constraints shaped the final direction of SmolPC 2.0. Rather than remaining a LibreOffice-first prototype, the project evolved into a unified offline AI platform with coding support as the flagship workflow and shared support for writing, presentations, image editing, and 3D workflows. This shift also changed the technical requirements: Windows-first packaged deployment, local privacy, easier setup, reduced installation burden, and hardware-aware local inference became core project goals rather than secondary concerns.

Project Goals

The overarching goal of SmolPC 2.0 is to deliver a free, offline, open-source AI assistant suite that integrates directly into the software students already use, including coding, LibreOffice, GIMP, and Blender workflows, while remaining realistic to deploy on a range of school and personal devices. Rather than assuming one fixed hardware profile, the system is designed to recognise the best supported local runtime path on the current machine and keep local AI practical without cloud dependence.

Specifically, SmolPC 2.0 aims to:

Gathering Requirements

Requirements were gathered iteratively rather than fixed once at the start. Because SmolPC evolved through several architectural pivots during development, the final requirements reflect both early stakeholder goals and constraints discovered through implementation.

1. Legacy baseline: we reviewed the original SmolPC LibreOffice and Ollama-based prototype to identify what was reusable and what prevented broader adoption.
2. Scope expansion: new project objectives introduced a dedicated coding workflow together with assistants for GIMP and Blender, turning the project from a single-application prototype into a multi-workflow platform.
3. Prototype-driven refinement: early assistant development exposed problems around duplicated inference logic, security boundaries, Windows packaging, and the deployment overhead of relying on a separately managed model server.
4. Deployment-driven refinement: as the goal shifted towards realistic school deployment, requirements expanded to include a unified app, a shared local engine, hardware-aware runtime selection, direct model management inside the product, and a setup flow that avoids asking students or teachers to install Ollama or manage inference infrastructure manually.

We did not rely on a formal pupil survey. Direct access to school-age users was limited for safeguarding reasons, so requirements were refined through supervisor discussions, teacher-facing feedback, prototype evaluation, and iterative internal testing of realistic workflows.

We also compared mainstream cloud AI tools with the deployment assumptions required for SmolPC 2.0. This was not intended as a general product ranking; it was a way to highlight why privacy, offline use, easier deployment, local control, and predictable classroom setup became decisive requirements for this project.

Later accessibility-oriented implementation work on voice interaction reinforced the same conclusion: if speech input or read-aloud support was included, it still needed to run locally, avoid separate cloud services, and degrade safely when audio hardware or bundled voice resources were unavailable.

Table 1. Comparison of deployment assumptions between mainstream cloud AI tools and SmolPC v2.0.

Personas

Based on the information we've collected, we constructed the following personas:

← →

Persona of a Student

A secondary-school student who uses a school or personal laptop for programming, coursework, presentations, and creative assignments. They are comfortable asking questions in plain English, but not configuring local AI tooling or debugging installation problems. They need a tool that feels approachable, keeps their work private, and helps across more than one subject area.

Persona of a Teacher

A teacher or teaching assistant who prepares lesson material, supports students in computing and digital media work, and values predictable deployment. They want useful AI support for classroom tasks, but cannot depend on cloud services, paid subscriptions, or setup steps that would be difficult to repeat across many student devices.

Use Case Diagram

SmolPC 2.0 — User Interactions

Figure 1. High-level use case diagram for SmolPC 2.0 as a unified local AI platform.

Use Case List

The following use cases describe how students and teachers interact with SmolPC v2.0 as a unified local AI platform. Each use case maps to one or more functional requirements listed in the MoSCoW section below.

MoSCoW Requirements

Requirements were prioritised using the MoSCoW method, separating what the system must deliver from what would be beneficial but is not critical. The list below reflects the final agreed requirements following stakeholder interviews and pseudo-user testing.

Functional Requirements

Non-Functional Requirements

References

1. UNESCO, "UNESCO spotlights how digital learning can promote equity in low-resource contexts", 31st March 2025, Available: https://www.unesco.org/en/articles/unesco-spotlights-how-digital-learning-can-promote-equity-low-resource-contexts
2. National Education Union, "Working lives of teachers and leaders report shows staff overwhelmed by workload", 27th November 2025, Available: https://neu.org.uk/latest/press-releases/working-lives-teachers-and-leaders-report-shows-staff-overwhelmed-workload
3. The Sutton Trust, "State schools falling behind in new AI digital divide", 16th July 2025, Available: https://www.suttontrust.com/news-opinion/all-news-opinion/state-schools-falling-behind-in-new-ai-digital-divide/