Standpoint — AI Crowd Experience Simulator

Organizers sell tickets to views they've never seen. Sketch your venue + add photos — AI builds a 3D sim to check any seat's sightline, safety, sound, and lighting.

Comment

Inspiration

I run Flip, an event ticketing platform in Vietnam (39K+ tickets sold, 21 organizers). Before that, I spent years at Meta building event ticketing infrastructure with Ticketmaster and Eventbrite integrations. One problem kept showing up on both sides of the market: organizers sell thousands of tickets to views they've never seen.

Ticketmaster has a venue simulator — but only for major arenas with CAD files and enterprise budgets. The mid-market organizer running a 3,000-person music festival in a converted warehouse has nothing. They find out about blocked sightlines, dead audio zones, and dangerous crowd pinch points during the event, from angry attendees.

What it does

Standpoint turns a rough sketch into an explorable 3D venue in minutes:

  1. Draw your layout — stage, seating blocks, GA zones, pillars, bars, exits, on a simple 2D canvas.
  2. Add context — upload photos of the space and describe it ("warehouse, 8m ceilings, stage on north wall"). GPT-5.6 interprets the sketch + photos and generates a parametric 3D scene graph: dimensions, materials, speaker positions, lighting rig.
  3. Sit anywhere — click any seat or standing spot to get a first-person view with a simulated crowd. Check what an attendee actually sees, raycast against heads and pillars.

On top of the 3D scene, Standpoint overlays:

  • Sightline scoring — % of stage visible per seat, obstruction detection
  • Safety heatmaps — crowd density per zone and egress distance to nearest exit
  • Audio coverage — speaker coverage cones and dead-zone detection
  • Lighting preview — rig placement generated by the AI scene graph

You can also just ask: "Where's the best cheap standing spot for sound and sightline?" — the model answers grounded in the computed scores, not vibes.

How we built it

  • Codex drove the entire build — we used it to scaffold the Next.js app, generate the three.js rendering pipeline, and iterate on the 2D canvas editor. [Add: specific Codex workflows you used — agent tasks, refactors, test generation]
  • GPT-5.6 is the venue architect: it takes the sketch (as structured JSON) + venue photos + text description and outputs a validated scene-graph JSON that the deterministic renderer consumes. AI decides what the space is; three.js decides what it looks like.
  • Stack: Next.js + TypeScript, three.js (instanced meshes for crowds, raycasting for sightlines), react-konva for the 2D editor, structured outputs for the scene graph.

Challenges we ran into

[Fill with real ones — e.g., getting GPT-5.6 to output spatially consistent scene graphs, tuning instanced crowd rendering performance, mapping photo perspective to real dimensions]

Accomplishments we're proud of

Photo + sketch → walkable 3D venue with zero CAD, zero 3D skills required. What takes an enterprise venue team weeks, an independent organizer can do before their morning coffee.

What we learned

[Fill post-build]

What's next

Piloting with real organizers on Flip's platform next month. Longer-term: embed the seat-view directly in ticket purchase pages, so buyers see exactly what they're paying for — and organizers can price seats by simulated experience quality, not guesswork.

Built With

Share this project:

Updates