By Caitlin Wilson and Barbara McGillivray
On 16 September 2026, researchers, legal professionals, UX designers, heritage practitioners, and members of the public gathered at King’s College London for a day-long hackathon. The goal of the day was to take real tools built on real collections, find the problems users have with them, and design something better together.
The event grew out of the first couple of rounds of user research conducted during the Lost for Words project (read more about it in this blog post), a collaborative doctoral award between King’s College London and The National Archives (TNA), for which Barbara McGillivray is the principal academic supervisor. In her doctoral project, Caitlin has been researching ways in which we can improve accessibility to the Find Case Law collection through semantic search. Through the development of prototype search engines, user research has been conducted to try to understand a few questions: How do people approach a complex collection like Find Case Law when they have little to no prior knowledge of case law? What do people actually do after they find a result? And how do we build a tool that a lay person can use, while still being useful to legal professionals? The hackathon was proposed as a research and knowledge exchange activity, bringing together people with different expertise and experiences to explore these questions and co-design potential solutions. This blog post shares the ideas that emerged from the day and highlights how collaborative events of this kind can help translate research into practical improvements for users.
In order to bring different needs into focus, we divided participants into three tracks, each centred on a different TNA collection and its associated tool. The first was Find Case Law — TNA’s service for searching and reading judgments from courts and tribunals in England and Wales — participants in this track worked with a semantic search prototype developed through Lost for Words that lets users search the collection using natural language rather than exact legal terms. The second was Legislation Chat which is an experimental tool that lets users ask questions about UK Legislation in plain English and receive answers drawn directly from the text of the law. And lastly, the Parliamentary Archives, which recently moved to its new permanent home at TNA (read more about that here), holds records of both Houses of Parliament; participants in this track worked with the Discovery catalogue and a set of records flagged during an internal project as containing potentially sensitive or outdated terminology.In the morning, participants worked through scenarios using the tools, noted where they ran into difficulty and chose a problem to address. And by mid-afternoon, groups were building solutions: some sketched interfaces, some wrote code, and others mapped user journeys or prepared design briefs. Everyone shared their work at an end-of-day show-and-tell.
What the groups built
The Find Case Law track produced two distinct prototypes. One group identified something we hadn’t named clearly before: the tool performs differently depending on how you search. Short, crisp queries like a case name or a short legal concept return good results in a way that’s already familiar from keyword search. Long, narrative queries like the kind a litigant might type when they’re describing a situation rather than naming a term work better semantically, but the interface doesn’t signal this. The proposed prototype differentiated between the two modes, expanding the search box for narrative queries, probing for clarifying details, and returning AI-generated explanations of why each case was relevant to the specific situation described. The second group focused entirely on what happens after search: the results page. Their design added a contextual sidebar pulling relevant guidance from GOV.UK, adapted dynamically to the search query. The point was to close the gap between finding a case and knowing what to do with it.
The Legislation group proposed a chatbot that would invite people to choose the role closest to their perspective — lawyer, researcher, small-business owner or member of the public — and then adapt its wording and its suggested next steps. In their live demonstration at the end of the day, the group entered the same question under four different roles. They reported how the model changed its language depending on the user with little extra prompting. The group also considered how a chatbot should handle trust and legal disclaimers. Its proposal was to explain the limits of the service, use GOV.UK visual cues, and end each conversation by directing users to a named legal charity for further advice.
The Parliamentary Archives track split into two groups, each tackling a different problem. One team built a Python-based prototype to flag potentially offensive or outdated terms in catalogue descriptions for review. It retrieved records through the TNA’s Discovery API, then checked the text in two stages: first, it looked for specified word patterns (regular expressions); second, it used a language model to sort terms into categories without examples labelled for this particular task (zero-shot classification). The prototype also scored terms for severity, historical context and confidence. It would not change catalogue records automatically: instead, it flagged records for staff review and drafted content notes for cataloguers to approve. The team also proposed a way for members of the public to report terms they encountered. The second group focused on external researchers and the language they use. For example, someone studying the decriminalisation of homosexuality may need to search for historical terms that researchers today might not think to use. The group proposed a search-expansion feature that would show related historical terms alongside results, let users choose which terms to include, and explain how the suggestions were generated.
Three findings
Looking across the discussions, three themes kept returning.
People come to collections with different needs. The hackathon groups did not identify a single interface that would work equally well for everyone, instead they all posited that interfaces and systems need to be adapted to different user needs. The Find Case Law teams, for example, proposed different ways to search and the Legislation group explored adapting responses to different user roles; while the Parliamentary Archives teams distinguished between the needs of internal staff and external researchers. A key question for development is therefore who the tool is for, and whose needs may otherwise be overlooked.
Finding a result is only half the job. Both Find Case Law groups and the Legislation Chat group identified the same gap: users arrive at a result and then don’t know what to do next, meaning that users may need guidance, support or help deciding what to do with a result. The groups saw limited signposting beyond the results they were working with. One Find Case Law team proposed a GOV.UK guidance sidebar and the Legislation Chat group proposed a referral to a legal charity where users could seek further advice. Helping users move from a result to a useful next step emerged as a common design priority.
Human review should be built in, not added as an afterthought. The Parliamentary Archives prototype was designed not to change catalogue records automatically: an archivist is required to review flagged records and approve any content notes. Other groups explored related safeguards, including explanations of why a case appeared, feedback options and routes to further advice. Across the proposals, oversight meant more than checking a system’s work; it also meant helping users understand the results and find further support.
How other heritage organisations could try this approach
A hackathon built around real collections and tasks can produce a different kind of output from a design sprint or user survey. In our experience, asking participants to try real tasks with existing tools brought specific points of difficulty into view. Asking groups to sketch or build a response then required them to make design choices, rather than offer only abstract recommendations. The result was a set of proposals that can be developed and tested at a further date; they are not evidence, by themselves, that those proposals will work.
We believe that many heritage organisations face a similar challenge. They may have extensive digital collections, yet still find it difficult to see where an interface does not meet users’ needs. Watching people try real tasks, and bringing together people who know the collections with people who use them, can help make those gaps clearer. A mixed group may also suggest approaches that no single group would reach on its own.
Taken together, the day showed the value of bringing people with different experiences together to work on real collections. The ideas developed are starting points, not finished solutions: they still need to be tested with the people who will use these tools. But one lesson was clear: improving access to digital collections starts with understanding what people are trying to do and involving them in shaping the tools that help them do it.
If you’re working on access and discovery for a heritage collection and want to know more about how we ran the day, we’d be happy to talk.
This event was made possible through two London Arts and Humanities Partnership initiatives: the Lost for Words collaborative doctoral project at King’s College London and The National Archives, funded through an LAHP Collaborative Doctoral Award secured by Barbara McGillivray in partnership with The National Archives, and a further LAHP Staff-led Activities Fund award supporting the hackathon.
