An AI-native events platform that books, routes and recovers tickets on its own

Approach

We framed Ticketdock as two jobs inside one app — a concierge for attendees trying to find something worth going to, and a co-pilot for organisers trying to get an event live before the weekend. The existing ticketing core was healthy, so we treated it as a system of record and wrapped an AI layer around it instead of replatforming. That meant drawing a hard line between deterministic flows (inventory, payments, seat holds) and probabilistic ones (intent parsing, recommendations, drafting), then picking the lightest model that could clear each job.

Solution

On the attendee side we shipped a natural-language discovery agent that takes a prompt like “Saturday night, indie, nothing seated” and resolves it against a vector index of live events, then hands the user a pre-filled checkout with seats already held. A recovery loop watches for stalls on the seat map, nudges with a contextual suggestion, and offers a one-tap alternative if the original section sold out mid-session. On the organiser side, an LLM-backed setup flow turns a rough brief into a draft event — pricing tiers, age rules, refund policy, social copy — which the organiser edits rather than authors. Payments, inventory and entry scanning stayed on the existing Node and Postgres stack; the agent layer sits beside it and is swappable per model provider.

What changed

Sessions that touched the AI assistant converted materially better than control, and organiser time-to-publish dropped from a multi-hour chore to a short review pass. More importantly, the product tone shifted — Ticketdock reads as an AI-native events platform now, not a ticketing form with a chatbot on top, and the same agent surface is what the roadmap for loyalty, resale and group bookings plugs into next.