Lessons from the Which? GenAI Hackathon
My name is Keir Kettle, Tech Lead for the Reviews and “Where To Buy” squad. I recently participated in the first GenAI hackathon at Which?. Though I’ve taken part in both company-organised and external hackathons before, this was my first since joining Which? four months ago. I have written this post to highlight the engineering culture at Which? and share some GenAI hackathon lessons and takeaways.
In the weeks leading up to the event, participants submitted and voted on ideas for hacks. Teams were then formed based on these votes and relevant skill sets or knowledge. One area for improvement in this process could be to collect problems rather than ideas, as I felt we missed some of the innovation that can come from this kind of collaborative process.
The proposed ideas ranged from the slightly over-ambitious, like “training an LLM”, to more straightforward ones, such as “summarising an article.” In my opinion, these ideas reflect our organisation’s current maturity in applying GenAI, as they are surface-level applications of LLM functionality. For the tech stack, teams were evenly split between AWS and Google Cloud.
Amit Vij (Director of Software Engineering) and team put in tremendous effort to organise the event and make it productive. External partners from AWS and Ancoris (our Google partner) attended to provide practical assistance to the teams. Environments and data access were provisioned; two floors of the office were set aside; and plenty of drinks and snacks were provided.
The idea I was assigned was “Which? Shorts” using Google Cloud tech stack. This aimed to condense advice articles down to their essence and display them as scrollable cards, similar to TikTok, Instagram Stories, or YouTube Shorts. Although my squad had considered a similar idea recently, we concluded there wasn’t enough user need, I was still keen to give it a shot — there were prizes for the top three teams, plus the chance for the winning team to be entered into a newly forming idea incubator.
The team makeup was interesting. Of the seven in the team, four were non-engineers: a product manager, a designer, a business improvement manager, and a user tester. I initially wondered how this would play out. Would these team members try to write some code? After all, isn’t the point of a hackathon to hack some code together and get a solution as quickly as possible?
What followed was a short lesson in user-centred design and the power of one-piece flow.
Within minutes we started — not by discussing the solution — but by understanding the problem. We examined how it fit with our knowledge of our customers and previous user research, identifying which parts of that research could inform the design and clarify our direction. Within a couple of hours, we had a problem statement that led to a design and a Figma prototype. The user researchers on the team could then use this prototype for actual user testing and the engineers could start working on the right solution — all before lunch!
This process was a great demonstration of the power of one-piece flow and its ability to focus alignment to a shared goal. While it seems counter intuitive but without context-switching and communication barriers — work flows efficiently. It was great to try it. Henrik Kniberg did a great talk on this and you can see the video on YouTube.
From a engineering point of view, we wrote some basic code in Typescript to get the contents of specific articles via a simple fetch, then used unfluff to extract the content and pass it to Gemini via the Google Cloud Vertex AI SDK. We made use of the prompt engineering library in Vertex Studio and quickly iterated on prompts. We landed on a one-shot instruction based prompt — which gave a high-quality output using the Gemini 1.5 Pro model. If I’m honest there wasn’t a lot of engineering required, testament to ease of implementing these types of technologies.
The following day we presented our hacks to a panel of judges, including an external judge (Barney Ajetunmobi, ex-product VP of Trustpilot). While, in my opinion, we did a great job of presenting our idea, the business case, and how it aligned with our organisational OKRs, we didn’t secure a place in the top three. The hackathon was won by a fantastic hack that aimed to analyse data from product testing quickly — a genuinely great use of Gen AI that promised to bring a little more joy to people’s working day.
In conclusion, this hackathon was a valuable experience — the combination of cross-disciplinary collaboration, time constraints, and competitive spirit was inspiring and something that is difficult to replicate in a traditional working environment. It highlighted the importance of user-centred design, the benefits of WIP limits, and the potential of Gen AI applications. It was a great opportunity to learn, collaborate, and innovate, and I look forward to applying these lessons in future projects.
Which? is the UK’s consumer champion, here to make life simpler, fairer and safer for everyone. As an organisation we’re not for profit and all for making consumers more powerful. Read more about Which? on our website.
If you’re interested in championing consumers with us, you can check out our open job vacancies on our careers site.
