Gallery
Description
https://makebetter-events.vercel.app/
Make Better Events — project submission
Build the right event before building it.
Elevator pitch
Make Better Events is an AI event strategist. Describe a conference activation: your goals, audience, budget, and constraints. It returns a strategic opportunity map, three genuinely different event concepts, a weighted scorecard, an editable production plan, and a skeptical pressure test. It helps you decide which event is worth producing, before anyone books a venue.
Live demo: https://makebetter-events.vercel.app · VIDEO DEMO: https://www.loom.com/share/baf152d4e5554f5b92db52b0882a0c94****
Inspiration
Almost every event tool starts after the most important decision is already made. Registration, ticketing, venues, check-in, guest lists — they all assume you already know what you're producing. The hard part comes earlier: should this event exist, who is it really for, and what would make someone choose it over the ten other side events that week?
That strategic judgment usually lives inside experienced producers and agencies. Meanwhile, the people increasingly asked to host — community ambassadors, ecosystem leads, field marketers — often get a budget without ever getting strategic training. Make Better Events puts a senior event strategist in the room for all of them.
What it does
You give it a messy, real brief — "I have $1,000 to activate my community during SF Tech Week" or "we're a CMO team weighing two side events at Korea Blockchain Week." It returns, in one flow:
- A Conference Opportunity Map — who's actually there, what they need, oversaturated formats to avoid, the strategic gap you could own, and a blunt "Should you host this at all?" recommendation (it can tell you to partner, go smaller, or not host).
- Three genuinely different strategies — divergent by mechanism, not just theme, each with its strategic posture, best-for, tradeoff, and the way it could fail.
- A weighted scorecard — eight dimensions with fixed weights and a written rationale per score, so the recommendation is earned, not vibes.
- A Feasibility Check — categorized warnings (budget pressure, staffing, capacity, measurement gaps) with concrete adjustments.
- An editable Strategy Board — the chosen concept expanded into guest journey, run-of-day, budget, risks, and measurement. For multi-event briefs it produces a full two-event portfolio with a combined budget.
- Event Vision — for each concept, the engine imagines how the room looks and feels (palette, materials, light, spatial flow, a signature moment), rendered as a designed scene.
- A Pressure Test — a skeptical senior producer critiques the idea, names the risks, and proposes the simplest version that still works.
No account. It runs instantly in the browser, and you can bring your own API key.
How we built it
Next.js (App Router) + TypeScript + Tailwind, deployed on Vercel. The intelligence is Anthropic's Claude, called server-side through a single /api/generate route with three modes — strategy, board, pressure — exactly three calls in the happy path. Every response is forced into a JSON schema via structured outputs and validated with zod (one retry, then a graceful, human error).
Design decisions that carried the product: the three strategies are generated in one call so they're aware of each other and diverge; scorecard weights are fixed in config and the weighted total is computed client-side, so the numbers are fair and never silently re-roll; the board is a discriminated union (SingleEventBoard | PortfolioBoard) so multi-event portfolios are a real capability, not a mockup; and Event Vision renders entirely in CSS/SVG from structured fields — no image model, no cost, no latency. State lives in localStorage (no database), with clearly labeled live / demo / fallback result states so a cached example is never mistaken for your real brief.
Challenges we ran into
- Making three strategies actually different. The default failure mode is three flavors of "networking mixer." We solved it with a divergence-by-mechanism rule and a single-call generation.
- Making the scorecard trustworthy. LLM-assigned weights are arbitrary and unstable, so we fixed the weights, computed totals in the app, froze the result, and labeled it "a comparison, not a guarantee."
- Multi-event portfolios under structured outputs. Representing two connected events cleanly meant a discriminated schema rather than optional-field soup.
- "Wow" without a photo model. Event Vision proves a designed CSS render can carry the feeling — instantly and for free.
- Instant, no-account use. A keyless public AI app needs rate limiting, cached fixtures, and honest fallback UX so it stays usable and never fakes a result.
Accomplishments we're proud of
Anyone can reach a full, scored strategy in under a minute with no sign-up. The product doesn't just generate output — it exercises judgment: it will argue against your own idea, simplify an overbuilt one, and tell you not to host. And it scales from a $1,000 ambassador dinner to a $75–100k executive portfolio with the same strategic questions. Lighthouse accessibility: 100.
What we learned
Structured outputs plus schema validation turn an LLM from a text generator into a dependable product surface. The prompts are the real IP — the difference between "AI wrote an event brief" and advice that sounds like a senior producer. And trust is a design problem: fixed weights, frozen results, honest fallback labels, and "this is a comparison, not a guarantee" are what make people believe the recommendation.
What's next
Conference research (pulling real context from a link), team collaboration and shared workspaces, learning from real event outcomes to build historical playbooks, and an optional photoreal toggle for Event Vision.
Built with
Next.js · TypeScript · React · Tailwind CSS · Anthropic Claude (structured outputs) · zod · Vercel