Ex-Google PM Vibe-Coded Palantir in a Weekend: Inside the Viral World View Globe

The setup is simple: a former Google Maps PM posted a weekend project and woke up to it trending. The project is called World View, framed as what you get if Google Earth and Palantir had a baby — live satellite positions, commercial and military flights, street cameras, traffic and seismic events fused onto a 3D globe with a classified-system skin. The whole thing was assembled in about three days with a crew of parallel AI agents.

The base layer is the full planet rendered with Google 3D tiles, a dataset the author once helped ship. On top sits a look system: switchable CRT, night-vision and thermal-style modes with controls for sensitivity and pixelation, plus browser-side post effects like bloom and sharpening. The point is practical as well as stylish — camera moves can be planned and recorded straight from the browser instead of faked later in an editor.

Navigation is built around presets. Cities and landmarks are resolved against OpenStreetMap 3D volumes so the camera centers exactly on the point of interest, avoiding the offset you get from raw coordinates. Shortcuts jump between viewpoints, which turns the globe into something closer to a shot list for content work.

Space is the first live layer. Every tracked object currently in orbit can be shown, with sparse or full detection views down to individual catalog IDs. Selecting one draws its path around the planet, so orbit type reads directly off the visualization rather than from a table.

Aviation comes next through the OpenSky feed: around sixty-seven hundred live flights appear on the globe at once. A detection view lets the presenter hop into any single aircraft and follow it, producing the spy-thriller framing the video leans on. The argument is that this kind of sequence no longer needs offline rendering — it happens live in the page.

Military traffic gets its own layer via crowdsourced ADS-B aggregators, shown in orange precisely because these flights are usually absent from mainstream trackers. Filtering isolates them, and the demo flies over the Pentagon area to show nearby military movements alongside satellites overhead. Stacked with commercial traffic, the author jokes, you get a personal panopticon built only from open feeds.

Ground detail is simulated rather than streamed. Road geometry from OpenStreetMap drives a particle system that imitates city traffic, demonstrated over London with sparse labels so the whole city stays readable, especially under night-vision and thermal looks. The same approach covers places without Google 3D coverage: Dubai works from oblique satellite imagery shot off-nadir, with the road network still rendering cleanly and the Burj Khalifa recognizable.

City cameras are real but slower: open feeds from Austin are pulled roughly one frame per minute and projected onto 3D geometry, so a crosswalk and passing vehicles visibly update when you jump between viewpoints. A more precise calibration pass is described as in progress. Seismic and earthquake markers stack on top, and the presenter stresses this is only a few days of layering with more streams on the roadmap.

The build method matters as much as the result. Several current assistants were used interchangeably through terminal CLIs, with a multi-terminal runner hosting three or four sessions and up to eight agents at once — one crew on shaders and style presets, another on data integration and particle systems. When the full traffic view crashed the browser, the fix was loading road classes in sequence, main roads first. The closing message is aimed at specialists: with domain knowledge plus agents that accept screenshots and plain instructions, this is the moment to build it yourself.

The strongest objection comes from the Palantir side: stitching open feeds together is not the same as fusing proprietary, access-controlled records with entity resolution and audit trails. Steelmanned, that critique is fair — this demo impresses on the visualization layer while the hard part of real intelligence platforms, permissioned joins across messy private datasets, is nowhere in scope.

What the demo does not show is cost and fragility. Thousands of aircraft plus every tracked object in orbit plus a particle system for road traffic pushes a browser hard; the sequential road-loading fix is presented as a clever trick, but it signals a ceiling. Data freshness varies too: a one-frame-per-minute camera, volunteer receiver gaps in crowdsourced tracking, and coverage holes in open flight feeds all need independent verification before anyone leans on them.

I treat the headline numbers as snapshots, not specs: flight counts and satellite totals change by the minute, and coding-assistant versions already moved during editing. The author's Maps background also shapes the story toward 3D tiles, so I cross-checked the stack against independent docs — the OpenSky API reference, the Maps Platform tiles FAQ, Cesium guides, and reporting on open flight tracking and vibe-coding security — rather than taking the narration at face value.

My judgment, in first person: for learning, prototyping, and content production this is a genuine path — a domain expert can now direct agents instead of hiring a graphics team. For operational or privacy-sensitive use it is not: I would not run unreviewed agent-written data plumbing near anything critical, nor point open camera and military-flight layers at real decisions without review. Prototype freely, verify before trusting.

vibe coding · 3d globe · live data