#GPUI is a great emerging framework which did not have maps component. So, we developed it from scratch for ourselves and now publishing this to open source.
gpui-maps.swatmobility.com/
There are also binaries of the agentic-oriented studio maps app for Mac which I've been using to generate maps for #30DayMapChallenge
#30DayMapChallenge Day 12: Power
Simplest idea about power map is to show the power grid at night. Since OpenStreetMap volunteers have mapped thousands of line sections, we will just use it, glowing by voltage over NASA's picture of the Earth's night lights.
Lowest voltages switch on first, the 765 kV and ultra-high-voltage DC lines last and brightest. Then a tour: Europe's lines running in step at 50 Hz from Portugal to Turkey; Egypt, where the grid follows the Nile; India, one synchronous grid since 2013; China's 800 and 1,100 kV lines carrying power thousands of kilometres east; and Africa south of the Sahara, where the lines thin out and many people still live without electricity.
Data: power lines © #OpenStreetMap contributors, via Overture Maps · Predicted grid: Gridfinder, Arderne et al. 2020 (CC BY 4.0) · Night lights: NASA Black Marble 2016 · Map: #OpenFreeMap, © #OpenMapTiles. Engine: gpui-maps
1
8
301
#30DayMapChallenge Day 10: Prompting only. And we may call it also "AI is the new UI" (for maps). And what can be more dynamic than city transit.
As day 10 requires building a map with prompting only, in the video below we can see how prompting with a specialised transit-oriented harness can automatically perform necessary actions to work with maps and GTFS data, then find most busiest routes, show most busiest bus stops and map of most loaded streets/corridors. It can answer even more complicated questions and analyze data on the fly.
We used to see tens of buttons and checkboxes in legacy GIS / transit analytics systems. This situation is rapidly changing now in agentic-first public transport analytics systems.
2
3
236
#30DayMapChallenge
Day 9: Urban-rural.
Under urbanization we usually mean migration of people from rural areas to cities and towns. But how long does it take to get to the city or nearest town from some rural area. Lets try to visualize it on the map. So the map is showing exactly this distribution. Blue is under 30 minutes, magenta is more than 6 hours.
So, actually we can notice that rural is not the same as remote. For example Bangladesh is mostly countryside, yet almost everyone there is within an hour of a city: towns are everywhere and the roads reach them. In the Ethiopian highlands, dense farmland sits hours from the nearest city. In much of Madagascar the trip takes half a day, and in Papua New Guinea many villages have no road at all.
Under the hood: millions of population cells joined with a global travel-time surface, and every map tile is one @ClickHouseDB query that returns a finished vector tile. Drawn on the GPU by our Rust map engine, on a globe. Data: travel time to cities, Nelson 2019 (2015 road network, CC BY 4.0) · cities: GeoNames (CC BY 4.0) · GHS-POP R2023A, European Commission JRC (CC BY 4.0) · Map: OpenFreeMap, © OpenStreetMap contributors
#30DayMapChallenge #ClickHouse #Rust #GIS #DataViz #Logistics
5
186
#30DayMapChallenge Day 7: 10 minute map
10 minutes for map generation is so short, that probably it was intended to be agentic. So, we can only observe how agent, tuned for GIS operations can generate map and interactions within a short amount of time. Looks like principle for new generation of GIS systems - No buttons, no UI. AI is the UI The chat is the whole interface.
The video speeds up the part where the agent is working and slows the flights down. Everything on screen, from the heatmap to the camera, is drawn on the GPU by our new Rust map engine that we recently started to integrate into applcations. Data: USGS Earthquake Hazards Program, all_month feed.
If you work with maps, gis, transit, vehicle routing problems, feel free to ping me
5
372
Modern public transportation analysis tools, for sure, should be efficient, precise and agentic (and written in @rustlang ).
GTFS, as a standard of data interchange between agencies allows to build generic tools that will work out of the box globally.
Industrial knowledge, accumulated over various public transit projects, give us ability to train and equip agents to extract this data not only precise, but also fast and in a convenient UI. Here is example of the interface for GTFS analysis. Besides standard reports, tool can work agentically with transit network as well as combine 3rd party datasets.
Well, and of course, buses should move... fast
Feel free to ping me if you analyse public transit networks in cities and look for such configurable open tools.
#GTFS #transit #maps #agentic #analysis #Rust
1
6
166
#30DayMapChallenge Day 6: Vintage
Modern map of MRT/Railways of Singapore is very well-connected, but if we speak about "Vintage" topic, then closer look deserves visualisation of the railway roads that are no longer in use. OpenStreetMap still remembers those roads.
Rendered without any external map depdencies on a rust-based map engine which been developing agentically for the last 4 months. Ping me if you are working with maps, struggling with big data and interested to try such map engine in your projects.
#30DayMapChallenge #Singapore #Rust #GIS #DataViz #Railways
#30DayMapChallenge Day 5: Senses: Sight
What can you see from the top of Mount Rainier?
Gold is ground in view; dark is ground hidden behind a ridge. We cast 1,440 sight lines from the eye over the USGS 3DEP elevation model, one every quarter of a degree, out to 230 km. The ground drops away with the curve of the Earth (less standard refraction).
From the summit, 4,392 m up, you see about 13,900 km², and the farthest visible ground is 195 km away. From Tolmie Peak fire lookout you see 4,900 km². Paradise, the visitor centre on the mountain's south side, sees only its own valley: 18 km². From the top of the Space Needle in Seattle you see 1,760 km² of city, water and hills.
The contours (every 100 m) and the sight lines are drawn on the GPU by @swatmobility new Rust map engine, and the video is rendered from a story script. No GIS tool was needed, and nothing is pre-rendered.
Data: USGS 3D Elevation Program, 1 arc-second DEM (public domain) · Lookout positions: © OpenStreetMap contributors · Map: OpenFreeMap, © OpenStreetMap contributors
#30DayMapChallenge #Rust #GIS #DataViz #Viewshed
1
16
411
#30DayMapChallenge Day 4: Clusters
Clustering a few thousand points on a map these days is easy. A lot of map libraries does it in the browser. But what about 8 billion people?
This is where @ClickHouseDB new MVT functions come in handy. Alexey Milovidov's team shipped ton of features recently and what is especially suitable here - new vector rile functions for maps: (MVTEncode, MVTEncodeGeom, MVTBoundingBox, which starts from ClickHouse 26.7).
Every tile on this map is one SQL query. ClickHouse groups 28 million population cells into clusters for that tile and that zoom, and returns the finished vector tile itself. It even fetches the basemap tile and merges it in.
ClickHouse is the tile server: the map asks for /{z}/{x}/{y}.mvt and gets one back in 0.1 to 0.4 seconds on a laptop.
Zoom in and new tiles mean new clusters: the globe, then South Asia, then Bengal. Past zoom 10 the clusters turn into points, one for every 0.7 km² cell, over Dhaka and Lagos. Then the Nile, Europe, and back out. Drawn on the GPU by new Rust map engine, on a globe. Nothing pre-rendered, no tile server, no clustering in the app. Day 1 showed the same 8 billion people as dots. Today same data but with the clustering.
Data: GHS-POP R2023A, European Commission JRC (CC BY 4.0) · Map: OpenFreeMap, © OpenStreetMap contributors. #30DayMapChallenge #ClickHouse #Rust #GIS #DataViz #VectorTiles #swatmobility
#30DayMapChallenge Day 3: Polygons
224,390 polygons: every US census block group, coloured by median household income
The flight goes between US cities. All polygons are drawn on the GPU by new Rust map engine with no external dependencies.
Data: American Community Survey 2023 5-year estimates and TIGER/Line 2023, US Census Bureau · Map: OpenFreeMap, © OpenStreetMap contributors
3
187
#30DayMapChallenge Day 2: Lines
(And a little bit about people mobility analysis and visualisations)
680,000 home-to-work commutes across England and Wales, as lines.
If you work with maps, you could have probably seen these flows as 3D arcs in tools like kepler.gl. The video starts there, pink where people live and blue where they work, with London pulling in commuters from half the country. What's new here: the same flows bundled into corridors, live, with two edge-bundling algorithms we implemented natively in our Rust map engine:
1. Edge-Path Bundling (Wallinger et al., IEEE TVCG 2022). Each flow is routed along a path through the other flows when that path is not much longer than the direct line. It is best for long flows between scattered places.
2. KDEEB, Kernel Density Estimation Edge Bundling (Hurter, Ersoy, Telea, EuroVis 2012). Each corridor climbs the density of all the others, so flows that run the same way merge into one trunk. It is best for dense local flows like commuting, and it is the one in the video.
Before bundling, flow ends are clustered into places, 40 km apart for the whole country and 5 km over a city, so the map re-bundles as you zoom in. Up close, London becomes a starburst of commuter lines. All 679,643 flows are drawn on the GPU and works blazingly fast (thanks to rust)
Prior work and references:
· Edge-Path Bundling reference code: github.com/mwallinger-tu/edg…
· Edge bundling of mobility flows (Digital Geography Lab, University of Helsinki): github.com/DigitalGeographyL…
· KDEEB paper: doi.org/10.1111/j.1467-8659.…
Data: 2011 Census origin-destination flows (WU03EW), Office for National Statistics, Open Government Licence v3.0 · Map: OpenFreeMap, © OpenStreetMap contributors
Feel free to contact me if you are also dealing with mobility analysis on a big scale
Why not participating in #30DayMapChallenge , especially when there are more tools to generate nice high-performant maps.
So, topic of the day 1: Points. And recently I saw Kyle Walker 's every-person dot map of 1.47B people in India and thought - why not putting this idea to the extreme - 8.19 billion people on the whole globe, with requirement to render it fast and smooth.
So, here is a short video - Tokyo, Bangkok, Singapore, Delhi, Dubai growing from 1975 to today, Yerevan, Tbilisi, Batumi, Istanbul, Moscow, Helsinki, Stockholm, Berlin, Paris, London, Madrid and New York, down to street level where you see the individual points.
Under the hood: the dots are aggregated live from @ClickHouseDB , 776 million population grid cells, and rendered on the GPU on a new map rendering engine, written from scratch in @rustlang ( @swatmobility maps )
Dots are drawn from the EU Joint Research Centre's GHS-POP R2023A 1km population grid (Global Human Settlement Layer, randomly placed within each cell.
Around a million points are on screen at any zoom, rebalanced as you fly in.
Data: GHS-POP R2023A, European Commission JRC (CC BY 4.0)
· Map: OpenFreeMap, © OpenStreetMap contributors
Feel free to contact me if you work with big data on maps especially!
4
313
So, it took three years since this tweet for AI to solve unsolved mathematical problem fully autonomously openai.com/index/model-dispr…
We may wake up one day and Riemann hypothesis or any other unproven hypothesis would be proven by AI or with AI help. Probably it's just a matter of time (GPT10?). Theoretical advances may also count.
169
Vibe-coded security nightmare is just starting. Will we see some AI cool-down soon for mission critical applications and subsystems?
We’ve identified a security incident that involved unauthorized access to certain internal Vercel systems, impacting a limited subset of customers. Please see our security bulletin:
vercel.com/kb/bulletin/verce…
109
Meanwhile my personal project for studying Georgian language is getting a bit better. In collaboration with very good Georgian translator and teacher, all AI-captioned texts are getting verified. The whole Atinati chapter is now verified, feel free to view --kartuligram.com/c/atinati__
150
weekend is a time for some experiments. introducing clawcast.tv - let your agent go live #buildinpublic
94



