When we set out to create Go With Pride, we had an ambitious goal: to build the UK’s most inclusive theatre and live events discovery platform for LGBTQ+ audiences. What began as an Innovate UK-funded prototype quickly evolved into a live platform integrating multiple data sources, cutting-edge generative AI, and an editorial sensibility shaped by deep community insight.
Here’s what we learned from building a recommendation engine with personality, an AI chatbot with attitude, and a cultural listing site powered by data, design, and intent.
1. Designing for discovery, not just search
We didn’t want users to have to "find" events — we wanted events to find them. That meant moving away from conventional filters and checkboxes, and towards a model where a user could have a conversation with an AI and receive contextually aware, creatively curated suggestions. We wanted wildcard results, not just drilldowns. That meant retraining ourselves to think about experience, not just information.
2. Why we used Perplexity alongside OpenAI
For show listings, we didn’t just need facts — we needed tone, nuance, and narrative. We used Perplexity as an AI layer to ingest show data (often from chaotic sources like PDFs, XML feeds, spreadsheets or scraped descriptions), and generate clean, consolidated summaries. This gave each listing a human-readable voice, without requiring full editorial overhead.
OpenAI’s GPT models powered our core chatbot and the Why Engine behind it. But Perplexity became our behind-the-scenes workhorse, cleaning up source material, merging show blurbs, and helping us extract the information we really needed: time, venue, tone, and vibe.
3. The hidden work: training data and prompt scaffolding
One of the most important (and underestimated) parts of the project was building the learning set to train and guide our chatbot. This wasn’t just a tech exercise — it was a content and UX process.
We worked hard to generate hundreds of structured entries across different types of shows, event formats, regional scenes, and tones. These entries were used to tune the tone and expectations of the bot, creating conversational patterns that felt helpful, non-patronising, and culturally informed. Without this learning set, the bot would have felt too generic to be useful.
4. Stitching together multiple ticketing APIs
The UK’s live events ecosystem is fragmented. We knew that if we wanted national coverage, we’d have to aggregate multiple ticketing feeds. We began with Ticketmaster and TodayTix, then added Skiddle to better support live music, club nights and regional listings.
Each API came with its own data format, rate limits, quirks and assumptions. Building our ingestion layer to standardise these feeds was one of the most technically complex and time-consuming parts of the project. But it was worth it. It allowed us to build a single, seamless user interface for users, even if the back-end was stitching together half a dozen messy sources.
5. The unexpected power of Google Image Search
Good recommendations need visuals. But getting high-quality, correctly licensed images for thousands of live events is a non-trivial problem. We used Google Image Search selectively, not just for presentation but to test associations and user context: what does the internet “see” when it sees this show title? This gave us insight into how mainstream or niche a production was, how it's been visually framed, and how to represent it back to the user.
6. Coaching the chatbot like a cultural critic
We learned quickly that getting the bot to "sound right" wasn’t just about technical prompting. It was about tone, empathy, and trust. Users don’t just want a recommendation engine — they want a trusted voice. Coaching the chatbot meant iterating on its tone of voice, injecting personality, and testing against edge cases. How does it respond to sarcasm? To a rude question? To a confused user?
It also meant providing enough data context that the bot could explain why it was recommending something — hence the Why Engine, a layer that pulls justifications and reasoning into the response. This was critical for trust and transparency.
7. Our biggest takeaway: cultural AI needs culture
The most important lesson we learned? You can’t build a truly helpful cultural AI product without cultural insight baked into every layer. The tech stack matters, but so does the training set, the tone, the curatorial eye and the communities you're speaking to.
Go With Pride isn’t just an events listing site. It’s a proof of concept for how AI can serve marginalised audiences better, by working harder to understand them. We’re proud of what we’ve built — and we’re just getting started.
8. Sharing ideas, building momentum
One of the most energising aspects of this project has been connecting with fellow Creative Catalyst innovators at events like the REMIX Summit at the Royal Academy and CIMIx in Austria. These gatherings have reminded us that the creative industries are hungry for smart, inclusive, technology-driven solutions.
We’ve been inspired by other teams working to bring audiences back to regional theatre, rethink the role of AI in live experiences, and reimagine access to the arts. There’s a real sense of shared purpose around using technology not just to optimise transactions, but to reinvigorate participation.
Collaboration across sectors and countries is going to be key. We’re already exploring opportunities to work with partners in Spain and Italy, and we see enormous potential in adapting this model to serve older audiences, live comedy lovers, and under-served music scenes. The tools exist. What matters now is how we use them.