Case Study · AI & Events

A live event that answers in every language

MainStage Pride draws a global crowd to Amsterdam, in many languages, all asking at once. We built the AI layer that answers them instantly, around the clock.

  • Client: MainStage Pride
  • Sector: Live Events
  • Services: AI Chatbot, Support Automation, Infrastructure

In the weeks around a major Pride event, attendees ask the same questions thousands of times: where, when, what is allowed, how to get a ticket. We designed and shipped a multilingual AI chatbot and the supporting infrastructure that routes those questions, answers them from a verified event knowledge base, and guides visitors toward the right ticket. The result is a support layer that scales with demand, holds a consistent voice, and lets the organising team focus on the event itself rather than the inbox.

Thousands of questions, one small team

A live event compresses demand into a few intense weeks. Questions arrive in Dutch, English and a dozen other languages, at every hour, across every channel, and most of them are variations on the same handful of topics. Handling that manually does not scale, and slow or inconsistent answers cost trust and tickets.

The organisers needed answers that were instant, accurate and on-brand, without growing a support team for a window that lasts only weeks. Accuracy was non-negotiable: event information has to be correct, because the wrong time or location at scale becomes a real problem on the day.

How the system is built

At the core is a retrieval-augmented chatbot built on large language models, grounded in a curated event knowledge base. Every answer is generated from verified source content rather than free recall, so the bot stays accurate as event details change. Around that core sits a multilingual routing and intent layer that decides what a visitor actually wants and how to respond.

  • Multilingual chatbot: A conversational assistant that answers in the visitor language, available 24/7 across the event window.
  • Event knowledge base: A structured, single source of truth for times, locations, rules and tickets that updates without retraining the model.
  • Support deflection: Automated handling of the high-volume, repetitive questions that would otherwise overwhelm a small team.
  • Ticket guidance: In-conversation routing that points visitors to the right ticket at the moment they show intent.
  • Human handoff: Clean escalation for edge cases and sensitive topics, so nuance always reaches a person.
  • Analytics dashboard: A live view of questions, intents and funnel behaviour that informs both content and operations.
  1. Discovery: We mapped the real questions visitors ask, the channels they use and the languages in play, then defined what the system must get right.
  2. Knowledge base: We structured event information into a verified, chunked and embedded source of truth ready for retrieval.
  3. Build: We engineered the RAG chatbot, multilingual routing, intent classification and ticket-funnel logic into one coherent system.
  4. Tune and test: We tested across languages and intents, tightened answers against the source content and refined the human handoff.
  5. Launch and observe: We shipped ahead of the event and used live instrumentation to watch demand and adjust content in real time.

The chatbot carried the weight of repetitive demand so the team could focus on the event. Outcomes are framed by what the system reliably delivers, not by figures we cannot verify.

Large Language Models · RAG · Next.js · TypeScript · Vector Database · Embeddings · Vercel · Edge Functions · Webhooks · Analytics