RoomForAll
RoomForAll is an AI-powered 3D venue accessibility, sightline, and emergency evacuation auditing and repair studio.
Core trick
Argon proposes a seating patch from first-person video, and deterministic 3D visibility checks verify it before a final reinspection.
Description
RoomForAll is an AI-powered 3D venue accessibility, sightline, and emergency evacuation auditing and repair studio. It addresses a pervasive blind spot in event and architectural planning: floorplans often treat attendees as uniform dots, overlooking how shorter seated viewers, wheelchair users, and guests needing extra legroom experience a space. These differences can reveal blocked sightlines, restricted turning space, cramped rows, and emergency egress bottlenecks.
This blind spot extends to AI-generated designs. We used Gemini Argon 4 to generate our initial venue views, but compelling visualizations can still overlook the perspectives and physical needs of the people occupying them. RoomForAll adds a verification layer: it examines those spaces from individual attendees’ viewpoints and checks whether the modeled geometry supports their needs. The same workflow supports auditing and repairing 3D models of actual architectural plans.
Users can load or upload a .glb venue, where specialized agents identify seats, tiered risers, presentation screens, and exit portals. The system records first-person video tours alongside a 2D social-force crowd evacuation simulation, exposing both individual viewing experiences and potential crowd-flow problems.
Powered by Gemini Argon 4’s multimodal video reasoning and independent, deterministic checks for 3D visibility and accessibility geometry, RoomForAll diagnoses obstructed views, inaccessible routes, and simulated evacuation congestion. It proposes bounded layout repairs, such as adjusting row spacing and relocating blocking equipment, then evaluates the revised layout against the same attendee requirements.
Before-and-after comparisons show what improved, which modeled checks passed, and what remains unresolved. RoomForAll helps teams critically evaluate AI-generated venue concepts and existing architectural designs, with verification tied to explicit modeled conditions. It supports planning decisions without claiming accessibility certification or guaranteeing real-world evacuation safety.
README excerpt
# RoomForAll — Existing Venue Accessibility and Viewing Review RoomForAll imports a prepared single-level venue layout, evaluates audience access, sightlines, and explicit comfort requirements across four audience profiles, records real first-person and overview inspection media, asks Gemini 4 Argon to diagnose obstructions and propose a bounded JSON seating/prop patch, verifies the candidate layout deterministically, and performs a final visual reinspection. --- ## 1. Parallel Workstream Ownership (`ARGON_MVP_PROMPT.md` §2) All four workstreams branch from this foundational commit and import shared schemas/constants exclusively from `src/shared/`. Only **Stream D** modifies root dependencies or `src/shared/`. | Stream | Exclusive Owned Paths | Primary Deliverables | Test Command | | --- | --- | --- | --- | | **A — Scene & Evidence** | `public/venue/`, `scripts/prepare-venue.*`, `src/scene/`, `tests/scene/` | Prepared venue asset/manifest (~24 places), `VenueViewport`, attendee cameras, `captureEvidence(snapshot, tourSpec)` | `npm run test:A` | | **B — Evaluation** | `src/evaluation/`, `tests/evaluation/` | `evaluateScene`, `applyAndValidatePatch`, `compareEvaluations` (15×9 visibility grid, routes, footprints, non-regression gates) | `npm run test:B` | | **C — Controller** | `src/server/`, `tests/server/` | Express API endpoints, Gemini/Argon provider adapter, real media upload, 110s deadline, run state persistence | `npm run test:C` | | **D — UI & Integration** | `src/