AI’s role in education: key takeaways from our January hackathon

In January, we teamed up with the Department for Education (DfE) and the Department for Science, Innovation and Technology (DSIT) to host a hackathon. This was our first event since the formation of the AI Education Content Store in September 2024.

The Content Store, designed to enhance learning for pupils and save teachers' time, will demonstrate how safely developed AI can improve public services. 

Over the past few months, the team has been busy designing and refining a pilot version. This includes:

• A prototype store of structured, tagged, and enriched educational data

• A prototype explorer app that allows edtech companies to browse and discover content

• A prototype API that enables integration of this data into edtech applications

The hackathon was an opportunity to see how generative AI and structured data could solve big challenges in education. We explored how early prototypes work in real-life situations. And discovered how the content store can help develop AI tools for education.

Co-designing solutions to real-life problems

The hackathon brought together edtech companies and the winners of the Department for Education’s AI Innovation Fund. They worked alongside AI and education experts from Faculty and the ImpactEd Group. We split everyone into teams and gave them problem statements to address, such as:

• Aligning formative assessments with UK Standards and Testing Agency (STA) specifications

• Making content easier to access for children with Special Educational Needs and Disabilities (SEND)

• Bridging gaps between US and UK curriculums, ensuring content is appropriately adapted

To kick off, we explored how generative AI models – like ChatGPT and Claude – could tackle these challenges. We then assessed their capabilities, testing how these models responded to educational prompts. We provided teams with access to the store. This included an initial data set of National Curriculum requirements and STA guidelines. They got to work developing prototypes and testing solutions.

At the end of the second day,  each team delivered a presentation. They shared what they built, the challenges faced and key lessons learnt.

Key takeways

The team uncovered important findings including:

1. Large language model (LLM) outputs vary in reliability. Using structured, reliable data, such as the National Curriculum and STA guidelines, can boost confidence in responses.

2. The public sector has a crucial role to play in making curated, structured, and tagged educational data available to edtech companies. Many attendees were unaware of the full extent of guidelines, frameworks and specifications set by the central government. They would have to invest significant resources to prepare them for use in their tools.

3. We need more data. The early-stage dataset provided valuable insights. But participants highlighted the need for more datasets. This includes exam specifications and examples of student work. AI models perform better when they have access to diverse, interconnected data sources. There’s potential in combining multiple educational resources. This will give AI a more holistic view of educator and student needs in particular contexts. 

4. Edtech companies need high quality tools for discovering and integrating educational data. As we were testing early prototypes, we only provided basic access to the data through an API. Observing how participants used the Content Store gave us useful feedback. It helped us understand usability, searchability, and what is needed for edtech applications.

What’s next?

We’ll apply our learnings to refine the Content Store, Explorer App, and API for better usability and functionality. At the same time, we’ll expand the dataset to cover a broader range of structured educational content.

Stay tuned for more insights from our hackathon at the end of March, where we’ll test the impact of these changes. We'll keep working with content providers, school educators and edtech companies to co-design the store.