Inspiration
Every underground music event, community festival, and independent show starts the same way — a voice note, a group chat, someone saying "we should do this." And then chaos.
Organizers spend hours — sometimes days — figuring out what services they actually need, which vendors are right for their vibe, how to write a professional outreach message, and how to explain their event to people who weren't in that group chat. Most of this work happens in spreadsheets, DMs, and guesswork.
We kept seeing the same pattern in the Vancouver underground and community events scene: incredible ideas dying in the logistics gap between having an idea and knowing how to execute it. The tools that exist — Eventbrite, vendor marketplaces, booking platforms — all assume you've already figured out what you need. None of them help you think.
That gap is where EventOps AI was born.
What it does
EventOps AI is a two-sided platform that connects event organizers with service providers through an AI-powered operations layer.
For organizers: You describe your event in plain language — the vibe, the size, the location, the budget. EventOps AI generates a structured event blueprint, builds a smart service checklist prioritised by importance (must-have, important, optional), allocates your budget across categories, surfaces operational risks specific to your event, and posts a public event brief that providers can respond to.
For providers: You build a profile once — your category, experience, price range, and the events you love working. You browse open event briefs filtered to your specialty. You submit a personalised quote directly to the organizer. You track every quote you've submitted and see when it's accepted or declined.
The AI layer: Every event blueprint is generated by GPT-4o from a plain-language description. Every service checklist is dynamically generated based on the specific event — not a generic template. Each quote gets an AI-powered fit assessment so organizers can make faster, better decisions. There's a contextual chat panel on every event that knows the full brief and can answer specific questions about budget, logistics, permits, and provider fit.
How we built it
Stack: Next.js 14 (App Router), TypeScript, Tailwind CSS, OpenAI GPT-4o, localStorage as a session and data store, deployed on Vercel.
Architecture: A single API route handles five distinct AI call types — blueprint generation, service checklist generation, vendor outreach, contextual event chat, and quote fit review. Each call type has a purpose-built system prompt tuned for that specific output. Blueprint calls use JSON mode to guarantee structured output that maps directly to the app's type system.
Data layer: We built a localStorage-backed store that simulates a real database — with seeded events and providers, cross-tab sync via the browser's storage event API, and a 3-second polling fallback. No backend, no database, fully functional across browser tabs.
Session system: Rather than building full authentication, we built a lightweight dual-session identity model. Organizer sessions and provider sessions coexist independently. When both are active, the app surfaces a role switcher so you can toggle between your organizer view and provider view on any event.
Design: We completely moved away from the original dark neon aesthetic to a minimal, professional design system — off-white surfaces, ink typography, subtle borders, clean cards. Built to feel like a tool, not a demo.
Challenges we ran into
The two-sided session problem. When someone acts as both an organizer and a provider in the same browser — which is exactly what happens in a demo — the app needs to know which role to show on a given page. We went through three iterations before landing on an explicit role switcher that lets you toggle between views on the event detail page. It sounds simple but the state logic underneath it is genuinely tricky.
Structured AI output at speed. GPT-4o's JSON mode is powerful but the service checklist prompt needed careful tuning. Early versions returned inconsistent structures — sometimes an array, sometimes a wrapped object, sometimes keys named differently. We built a normalisation layer in the API route to handle all variants without breaking the UI.
Cross-tab data sync without a backend. localStorage doesn't broadcast writes to other open tabs automatically — only the storage event fires across tabs, not within the same tab. We needed quotes submitted in Tab 2 to appear in Tab 1 without a manual refresh. The solution was combining the storage event listener with a 3-second polling interval, which gives real-time-feeling sync with zero infrastructure.
Scope discipline. The hardest challenge was deciding what not to build. The feature surface of a two-sided event platform is enormous — payments, calendaring, reviews, messaging, notifications. Every hour spent on a non-core feature was an hour not spent on the AI intelligence layer that actually differentiates us. We wrote "IS THIS THE INTELLIGENCE LAYER?" on a sticky note and checked every feature idea against it.
Accomplishments that we're proud of
The role switcher for dual sessions is something we haven't seen done cleanly in any hackathon project. It solves a real product problem — people naturally wear both hats — and it lets us demo the full two-sided flow in a single browser without any awkward context-switching.
The AI checklist generation is genuinely surprising. Describing "a rooftop corporate launch in Toronto for 80 people" produces a meaningfully different checklist from "an outdoor rave in Vancouver for 300 people." The priorities, reasons, and cost estimates are specific to the event — not generic. That's the core value proposition working exactly as intended.
The quote fit review feature — where an organizer clicks "AI review this quote" and gets a two-sentence honest assessment of provider fit — is something nobody in the event space has built. It's a small feature with outsized demo impact.
We shipped a full two-sided platform with real AI across five touchpoints, a working session system, cross-tab sync, and a clean minimal design — in a single hackathon.
What we learned
The intelligence layer is the product. We started thinking about EventOps AI as an event planning tool. We ended up understanding it as an operations intelligence layer — the thing that exists before the marketplace, before the booking, before the transaction. That reframe changed every product decision we made.
Prompts are architecture. The quality of the checklist, the blueprint, the outreach messages, and the quote reviews all depends entirely on the system prompts. We spent as much time refining prompts as writing UI code. A bad prompt produces generic output that breaks the product illusion immediately. A good prompt makes it feel like the AI actually understands your event.
Session state is underrated. Most hackathon projects skip identity entirely or bolt on full auth at the last minute. Building a lightweight dual-session model early meant the product actually felt like it had users — not just a demo state. The "you're the organizer / you're viewing as provider" banners changed how the app felt to use.
The underground scene doesn't want a marketplace. It wants a network. Every conversation we had while building this reinforced that community-driven events are built on trust and lineage — who you've worked with, who your friends vouched for. A rating system misses this. A portfolio of past events doesn't. That insight is the foundation of where this goes next.
What's next for EventOps AI
Vendor profile claiming with lineage. Providers already exist as matched suggestions. The next step is letting them claim their profile and build a public record of every event they've worked — not star ratings, but lineage. "Bassline Audio has worked with 12 organizers in the Vancouver scene" is more meaningful than 4.8 stars.
Event archive and remixing. Every event run through EventOps AI becomes an optional case study. Future organizers can find events like theirs, see the blueprint that worked, the vendors that showed up, the risks that mattered. And they can fork that blueprint as a starting point — GitHub for events.
Real database and auth. We deliberately built without a database for the hackathon. Supabase is one afternoon of work. When we add it, all the localStorage logic maps directly to database calls — we architected for this swap from the start.
City expansion. Vancouver is the seed. The underground and community scene exists in every city — Berlin, London, Melbourne, Mexico City, New York. The intelligence layer is city-agnostic. The vendor data isn't — but that's a growth problem, not a product problem.
The real vision: EventOps AI becomes the operating system for independent and community events. Not the marketplace. Not the ticketing platform. The layer of intelligence and trust that makes all of those things work better.
Built With
- next.js
Log in or sign up for Devpost to join the conversation.