CreatorsHub

InfraTwin-Live: a live 3D digital twin that connects the warnings before a bridge, tunnel or dam fails. THE PROBLEM Genoa 2018 (43 lives), Morbi 2022 (135 lives), Pittsburgh 2022: each collapse had warning signs on record. Nobody connected them. Structures are still inspected by hand every few years, sensor alarms drown in temperature noise, and alerts, photos, drone flights and repair decisions live in separate systems with no shared model of the structure. WHAT WE BUILT One product that runs Monitor → Inspect → Act on a live twin: - Twins on real structures: each asset is a 3D model of its members (piers, girders, lining rings, monoliths), coloured by live sensor health and anchored on the real structure from OpenStreetMap, shown in satellite view. - Real data: 15 months of monitoring from railway bridge KW51 (KU Leuven). A temperature-and-frost model removes environmental effects; the twin catches the bridge's real 2019 strengthening as a stiffness change, and Muse Spark assesses it like an engineer, streaming its reasoning. - AI Inspector: Muse Spark reads a real inspection photo (dacl10k) like a senior inspector, giving defects, severity and a written assessment. SAM 3.1 makes each defect pixel-accurate, measured and pinned to its member. On the sample shown it matched the expert labels: 2/2 exposed rebar, 1/1 spalling. - Seeing around obstacles: Muse Wonder calls the twin's geometry engine as tools, scores camera angles and plans a collision-free drone flight around a pipe hiding a crack: 30% → 100% visible. - Time-travel: Muse Spark reasons how the defect progresses; Muse Image shows it at 6 months, 2 years untreated, and repaired. - Hands-free: Muse Voice Transcribe + Muse Spark turn one spoken sentence into a field note and a new drone mission. - Global view: 638 real bridges, tunnels and dams (Wikidata) on a live globe with an operations dashboard: what's failing, repairs in progress, a live feed. META MODELS, ALL LIVE SAM 3.1 (box + point prompts per defect), Muse Spark 1.3 (vision, reasoning summaries, strict JSON-schema outputs), Muse Wonder (agentic function-tool loop), Muse Image (image edits), Muse Voice Transcribe (speech). Each model does a distinct job; together they turn a warning into an action in minutes instead of a site visit. Integrations for Muse Glimmer, Muse Spark Open and SAM 3D are written and await GPU deployment. BUILT LIKE A PRODUCT FastAPI + React, Three.js, MapLibre; WebSocket telemetry at 2 Hz; access-token auth, daily spend cap, retries with backoff; Docker and one-command start; tests; a self-running narrated demo; MIT licence. The README states exactly what is real and what is simulated. WHY IT MATTERS Millions of structures age faster than inspectors can visit them. InfraTwin-Live gives every owner a senior engineer's eyes on every structure, every day, so the next warning gets connected.