staniverse — media circuit about GenAI, XR and digital culture

The essence

staniverse is my own media circuit about GenAI, XR, automation and digital culture. It is not just a channel: Telegram is the live flow, YouTube carries long formats, and this static Astro site gives the material durable addresses, topics and relations.

How the site works

Astro ships finished pages, a graph JSON file and a search index. There is no language-model request from either the browser or the server. The repository holds typed collections for works, projects, articles, experiments and publications; a work remains a framed case, while a project remains a living system with status and a next step.

Relations are a typed graph rather than a paragraph about how things connect. An edge preserves its source, target, type, evidence, provenance, confidence and review state. A Telegram link can create an edge without merging posts; a thread only joins through replies, albums or explicit continuation markers. Topical similarity alone is not treated as causality.

The garden is the catalogue for the corpus: search, date sorting, list and map views, and filters for format, topic, channel, related project and media. Filters live in the URL, restore with Back and Forward, and combine OR within one dimension with AND across dimensions. The map distinguishes evidence-backed relations from topical recommendations.

Telegram archive and editorial control

A full Telegram Desktop export is the local baseline; Telethon and MTProto then fetch only messages newer than lastMessageId and merge them by message_id. Text, entities, reply chains, albums, links and media are normalised before pages are generated. Images become WebP, video H.264 MP4 up to 960 px, audio MP3, and the source archive stays outside Git while the optimised showcase remains in it.

Enrichment never rewrites an authored post. A job is keyed by id, text hash and prompt version. Topics and relations are separate passes; a proposed relation needs an exact evidence fragment, then a person accepts or rejects it. Pending and rejected proposals never appear as facts. A deterministic local layer keeps import working offline, while provider keys stay out of the client bundle.

Languages and universe

Russian is the authored original. Key routes — home, works, projects, articles, lab with experiments, about and contacts — have English twins through one route-pair table that also emits hreflang. Structural facts on English pages come from the Russian source entry. The archive, topics and universe are still Russian-only rather than being presented as unedited machine translation.

Universe is a full-screen map of the corpus: a selected star opens its context and live relations. It works with keyboard and pointer on desktop, touch and a virtual joystick on phones, and asks the browser about immersive-vr or immersive-ar support instead of assuming a device list. Camera access starts only after explicit AR entry and is released on exit; generative Web Audio also remains off until the visitor enables it.

Lab

Lab is a separate catalogue of runnable experiments, their launch URLs and concise mechanical breakdowns. Three 8th Wall AR demos and five Google XR Blocks pieces — REALITY//FIELD, WEATHER//ROOM, CITY//ORBIT, SOUND//SPACE and ECHO//ROOM — live in the open xr-experiments repository and are served from GitHub Pages. Google XR Blocks targets Quest 3: Safari on iOS can lack the required WebXR capabilities, so those experiences may not start there. Stanley Simulator, Parametrick and Craving keep their own launch addresses.

What has changed

The old published garden was replaced by a static Astro site with typed content and a shared relation graph.
The Telegram archive became a durable, incremental corpus with preserved authored text and media constraints.
Search, filters, graph map, bilingual route pairs, the Universe and the Lab now sit on the same content model.
New XR work stays isolated from the main site build while remaining discoverable and documented here.

Stack

Astro static build, TypeScript, typed collections, three.js, WebXR and Web Audio API
Python and Telethon over MTProto; TypeScript normalisation and materialisation; systemd scheduling and media audit
GitHub Actions and Yandex Object Storage delivery
Google XR Blocks, three.js, 8th Wall, A-Frame, Immersive Web SDK, p5.js and Next.js in the Lab
OBS, Unity, VRChat SDK and GenAI tools for production

Immediate focus

Bring the archive, topics and universe to the English surface without pretending that a machine translation is authored.
Run the full enrichment only after a reviewed 30-source pilot, then inspect proposed relations by hand.
Keep strengthening the topic vocabulary, long-lived articles and experiments that go beyond rendering.

Links

Original (Russian) →