
Ex-Google Maps PM Vibe Coded Palantir In a Weekend
What this covers
This is a ten-minute walkthrough of WorldView, a browser-based geospatial dashboard that Bilawal Sidhu, a former Google Maps product manager, built in about three days with parallel AI coding agents and public data feeds. It began as a viral post on X, where people said he had "vibe coded Palantir." A Palantir co-founder replied that it lacked real proprietary data fusion. Sidhu concedes the point, but treats the fact that the comparison came up at all as a sign of what current models can do. The video then goes through the layers one by one.
The layers sit on Google's 3D Tiles, which Sidhu helped launch at Google. They include every satellite in orbit, roughly 6,700 commercial flights from OpenSky, and military aircraft from the crowdsourced ADS-B Exchange. Simulated street traffic is drawn as a particle system over OpenStreetMap road data. The CCTV feeds come from public Austin traffic cameras that refresh once a minute, and a seismic layer shows earthquakes. Styling comes through CRT, night-vision and FLIR modes, with post-processing such as bloom, sharpening and LUT grading applied live in the browser.
Several of the more specific details concern problems he solved along the way. He centers camera presets on landmarks by querying the 3D volume instead of using latitude and longitude, which would misalign the view. For places like Dubai, which lack 3D scanning in Google Earth, he falls back on oblique satellite imagery. When the traffic particles crashed the browser, he had the agent load roads in sequence, main roads first.
The closing section covers the workflow. He runs three or four terminals controlling four to eight agents (Gemini 3.1, Claude 4.6, Codex), with one on shader styling, another on data integration and another on particles. He argues that building this in code is far easier than in Blender, and urges developers with domain expertise to build their own versions rather than wait for large organizations.
Developers can now rapidly prototype sophisticated geospatial intelligence dashboards by combining open-source data feeds (satellite tracking, flight data, CCTV, seismic activity) with AI-assisted coding, democratizing tools that previously required specialized expertise and months of development.
- Open-source intelligence APIs (real-time satellite data, OpenSky flight tracking, ADSB military planes, CCTV feeds) are freely available and can be integrated into browser-based visualizations
- AI coding agents (Gemini, Claude, Codex) can rapidly build complex geospatial applications when given domain expertise and creative problem-solving guidance, reducing development time from months to days
- The combination of these tools enables creation of 'personal panopticon' dashboards with real-time multi-layered situational awareness, previously only accessible to intelligence agencies and large tech companies
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
Real-time satellite tracking data showing all satellites currently in orbit is publicly available and can be integrated into browser-based visualizations with satellite IDs and orbital parameters.
“So this is all the satellites that are in orbit right freaking now. This is real-time, ladies and gentlemen.”
OpenSky provides real-time live flight data for approximately 6,700 commercial flights currently in operation that can be integrated into geospatial dashboards with tracking capabilities.
“Once we start zooming into the planet, we can also use OpenSky to load in live flight data. So give this a second to pop in. We've got 6.7K flights that have been loaded in.”
Google's 3D Tiles technology, launched while the speaker worked at Google, enables efficient rendering of massive 3D geospatial datasets in web browsers.
“You've got the full globe in here. 3D tiles, which by the way I had a chance to work on and launch at Google. Google's 3D tiles up in this joint”
AI agents can now democratize development of sophisticated geospatial applications, allowing developers with domain expertise (but not necessarily geospatial expertise) to rapidly prototype intelligence dashboards.
“Now you can do it, too. I mean, the AI isn't going to do it all for you. Like one of the issues I ran into when I was doing this panoptic view is like it would just spawn too many particles and just like immediately crash the browser.”
Developers should leverage their domain expertise (e.g., military tactics, urban planning, logistics) to build AI-assisted geospatial applications rather than waiting for large organizations to do so.
“But let me show you what's possible right now and encourage you to do the freaking same. Because if you got domain expertise, this is the time to put it to work.”
By combining open-source data feeds (satellites, military and commercial aircraft, CCTV cameras), you can create a 'personal panopticon' using only openly available data that provides comprehensive real-time operational awareness of a region.
“Then you throw in the military planes, then you throw in all the regular planes and oh my god, you have built yourself a personal panopticon all using open-source data feeds.”
The speaker can perfectly center camera views on OpenStreetMap landmarks by querying the 3D volume data rather than using latitude and longitude, which would result in misaligned positioning.
“And what's actually cool is that when you jump to these different points of interest, I'm actually looking at OpenStreetMap at the 3D volume of each of these points of interest. So they are perfectly centered. If you just try to do this naively with latitude and longitude, your camera may end up somewhere over here and not where you want it to be.”
A geospatial intelligence dashboard integrating real-time satellite tracking, live CCTV cameras, military and commercial flight data, and a 3D model of the world can be built in 3 days using AI agents.
“I built this in 3 days with an army of AI agents.”
ADSB, a crowdsourced open-source intelligence website, allows tracking of military aircraft that are typically not tracked on public commercial flight tracking websites.
“So there's this website called ADSB that essentially uses a bunch of crowdsourced data to track all the military planes that are on the planet right now. So all the orange stuff that you see right now are military planes that typically are not tracked on these plane tracking websites.”
AI coding agents (Gemini 3.1, Claude 4.6, Codex 5.2/5.3) can be used via CLI tools to parallelize multiple development tasks, with the speaker running 3-4 terminals simultaneously to control 4-8 different agents on different project components.
“If you're wondering which coding tool to use to make this stuff happen, all of them work. I used Gemini 3.1 and Claude 4.6 and even Codex 5.2. Actually, we're on Codex 5.3 now.”
One agent can be assigned to build shader styling, another to data integration, and another to particle system development for traffic and flight visualization, allowing parallel development of different components.
“For example, I had one agent building out all the shaders on top, right? The way this works is like you have some of the agents doing your style presets. Maybe you have another agent that's going and doing all the data integration. It builds your particle system for street traffic or military and flight data.”
Oblique satellite imagery (looking off-nadir) provides perspective view data that can be rendered even in places like Dubai where Google Earth lacks 3D scanning data.
“So for example, Dubai doesn't like, you know, US companies going and 3D scanning their stuff, this approach will still work. So that's the Burj Khalifa. Like this is an oblique imagery shot, right? So this is all off a satellite imagery that's looking off nadir.”
When particle system density became excessive and crashed the browser, the speaker resolved the issue by instructing the AI agent to implement sequential loading, prioritizing main roads before arterial roads.
“The AI isn't going to do it all for you. Like one of the issues I ran into when I was doing this panoptic view is like it would just spawn too many particles and just like immediately crash the browser. And so basically I told it, 'Hey, do like sequential loading. Why don't you load in all the main roads first and then get all the arterial roads.'”
Multiple data layers can be efficiently composed in browser-based geospatial tools by turning different data sources on and off, allowing users to isolate specific data types or combine them for comprehensive situational awareness.
“Like you take this, you throw the satellites on there, you throw in some of the military flights and like suddenly you've got a pretty freaking comprehensive picture of what's going on.”
Real-time CCTV camera feeds from public traffic cameras in Austin, Texas are openly available and can be integrated into geospatial dashboards, updating at one frame per minute.
“Essentially, what I've got going on is basically this whole panel that I created that will let me turn on CCTV cameras and pull from the live CCTV cams that we have in Austin, Texas. Now the cool thing about this is like this is all openly available.”
Compared to traditional video editing workflows, coding geospatial visualizations and screen-recording provides superior control and eliminates export-edit-import cycles.
“Like you could totally try and make something like this in VIO, but why not just code it up and screen record it? You get way more control this way.”
The speaker worked on Google Maps and has experience launching geospatial technology (Google 3D Tiles) at scale.
“I worked on Google Maps, right? I worked on this 3D tile data set that's out in the world.”
Developers previously faced a steep learning curve when working with 3D tile datasets like Google Maps, making it difficult for non-specialists to build with this technology.
“I worked on Google Maps, right? I worked on this 3D tile data set that's out in the world. And the biggest thing we'd run into is like developers would have such a steep learning curve to make all of this stuff operate.”
Building sophisticated geospatial visualizations in Blender would be 'order magnitude harder' than building them directly in code in a web browser with AI assistance.
“Doing something like this in Blender would have been order magnitude harder. Doing it directly in code in my browser, it's insane what you can do.”
Browser-based geospatial visualization tools allow applying visual filters and post-processing effects (bloom, sharpening, LUTs) in real-time without external video editing software.
“But if you want to also, you can use this to make some very, very cool visualizations. I mean, just look at this, ladies and gentlemen. Like why would I go try and build this in After Effects when I could do this in real-time in my browser”
The speaker's Palantir World View dashboard garnered significant social media attention, with the Palantir co-founder commenting that the project was missing 'real proprietary data fusion.'
“And the Palantir co-founder says, 'I'm missing real proprietary data fusion.'”
Seismic activity data is publicly available as a global data layer that can be integrated into geospatial dashboards to show earthquake locations and magnitudes.
“I've also got tags of every single earthquake, like basically seismic activity that's happening across the globe. So you can start layering this thing up, right?”
Mode switching (CRT, night vision, FLIR) with adjustable sensitivity and pixelation settings can be implemented in browser-based geospatial tools, providing multiple viewing modes for the same geospatial data.
“The first thing you can do is switch between a bunch of different modes. Essentially, if you click through these numbers, I can switch between CRT, night vision, and FLIR. And I've got full control over here. I can change like the sensitivity. I can talk about how much pixelation there is.”
Sparse labeling of road networks (showing fewer labels) provides better visualization of entire cities, especially in night vision and thermal imaging modes.
“I actually like doing just sparse labels cuz that way I can show you the whole freaking city. And especially when I go into things like NVG night vision mode and thermal mode, these things look spectacular.”
OpenStreetMap road network data can be queried to render simulated traffic particle systems that visualize city traffic patterns.
“So I've got in this case essentially using all the OpenStreetMap road network data, I can query it and essentially spawn a particle system that emulates city traffic.”
Camera calibration is the technical process required to align real-world CCTV camera feeds onto 3D geospatial models, and the speaker is developing a more sophisticated calibration system.
“I'm working on an even more sophisticated calibration system where I can drop a couple points and connect this. That's going to make this near perfect and figure out all the shader stuff that I need so that it looks very, very realistic projected on.”
Real-time vehicle tracking at urban scale is now possible by combining OpenStreetMap road networks with traffic particle systems, creating live city traffic visualizations.
“Now if suddenly I've got tracked points across the whole freaking London Bridge in the city of London, baby.”
The speaker mentions Grok (Elon Musk's AI model) summarized the project as 'Google PM builds spy thriller geospatial dashboard,' framing it as intelligence-related.
“Grok summarized it as X Google PM builds spy thriller geospatial dashboard.”
The speaker is a Google product manager based on references to working on Google Maps and Google 3D Tiles, though this is not explicitly confirmed.
“I worked on Google Maps, right? I worked on this 3D tile data set that's out in the world.”