Union Station, Los Angeles · Launching Soon

    Memory Space: ChinatownAR

    A geolocated WebAR experience that brings the oral and written histories of Los Angeles' original Chinatown to the ground where it stood.

    A Neighborhood Beneath Union Station

    Los Angeles' original Chinatown was displaced in the 1930s to build what is now Union Station. Much of the community's heritage survives in photographs, census records, and the memories of former residents and their descendants.

    Memory Space: ChinatownAR brings that heritage back to the site itself. Instead of scripted narration, visitors hold open-ended voice conversations with AI characters whose answers are grounded in the real archives. Your guide is Kongque (孔雀), an AI peacock who remembers the neighborhood.

    Why a peacock? Kongque takes her name and form from Peacock Alley, an Old Chinatown street named for the wild peacocks that wandered it. The peacock motif ran throughout Old Chinatown's architecture and cultural landscape, and carries venerable symbolism in Chinese folklore.

    Meet Kongque Now

    QR code that launches the Memory Space: ChinatownAR web AR experience
    1. 1Scan the QR code with your phone. Click open in App Clips.
    2. 2Allow camera and microphone access, then aim at the ground.
    3. 3Tap Begin Experience and say hello. 你好 works too!

    This QR code launches the build that works anywhere, so Kongque can appear on whatever ground is in front of you. At Union Station, the full experience uses an Immersal VPS scan of the station interior to anchor her to the real architecture where Old Chinatown stood.

    Kongque the AR peacock fanning her tail on a sunlit sidewalk, seen through a phone browser

    The Technology

    Augmented reality, live multilingual voice, and knowledge-grounded AI characters, all running in the browser on Estuary's infrastructure.

    Estuary Platform

    Estuary powers Kongque's voice interactions, knowledge bank, and action calling with fluid turn-taking and natural interruptions.

    Voice-Optimized RAG

    Kongque grounds her responses through low-latency retrieval-augmented generation across a knowledge bank of oral and written histories of Old Chiantown.

    Code-Switching Speech

    Soniox V4 handles multilingual, code-switching speech recognition, and ElevenLabs 2.5 Flash provides low-latency TTS, so Kongque understands mixed language input and pronounces Chinese names and terms correctly.

    Visual Understanding

    Multimodal interactions are powered by the Gemini 2.5 Flash vision language model so the characters can respond to what the visitor sees.

    WebAR with Mattercraft

    Built on Mattercraft with the Estuary Web SDK. The whole experience runs in a mobile browser from a single QR code. Compatible with iPhone and Android, no downloads required.

    Immersal VPS Anchoring

    Immersal Visual Positioning System is leveraged to spatially anchor Kongque in LA Union Station using an interior scan of the space.

    Want an AI Character for Your Museum, Venue, or City?

    Memory Space runs on the same Estuary platform available to every builder: real-time voice, knowledge bank RAG, persistent memory, and cross-platform SDKs. Start building today or reach out if you would like to collaborate!

    Start BuildingTalk to Us