Growing up, geography was never my forte. I had no sense of space (no pun intended), couldn't read a map to save my life, and somehow ended up spending plenty of time studying geographic science. Somewhere along the way, it clicked, and it's been my favourite rabbit hole ever since.
Now I work in GeoAI for cities, mostly across mobility, civic tools, and urban equity. Here you'll find things I've built for companies I've worked with, alongside a pile of analyses I did out of curiosity or the night before an assignment was due.1
NUST, Islamabad: Learning to Read the Ground
Geoinformatics is the art of looking at the Earth from above and not trusting the first answer.
My undergraduate degree at NUST was remote sensing, photogrammetry, geodesy and a great deal of GIS. I learned to pull land cover out of Sentinel-2 and terrain out of SRTM, and why a map's projection is never a neutral choice.
The project I remember best weighed seven criteria (slope, irradiance, land cover, distance to the grid) to site utility-scale solar in a Pakistani district, with AHP weights I had to defend line by line. It was the first time a map I made was meant to decide something.
2023: Maps That Have to Work on Monday
In my second year I started shipping maps for a company seven thousand miles away.
Adgile Media Group in New York runs fleets of mobile billboard trucks, and I joined remotely as a geospatial software engineer. I built the real-time side of their platform: live GPS streamed into PostGIS, raw traces snapped to roads with OSRM, and a React and Mapbox dashboard with live markers, route playback and fleet health. Lectures in Islamabad by day, New York stand-ups by night.
In my final semester Earthbound AI in San Francisco brought me on as well, so I graduated in June 2025 already working for two American companies on two different clocks. Real users taught me what coursework hadn't: a map that is slow, or wrong, is a map nobody opens twice.
Earthbound AI: Rivers and Pipelines
Where a pipeline crosses a river, the river is always trying to move.
After graduating I went full time on both jobs. At Earthbound I was the primary developer on the geospatial apps that pipeline operators use to watch that happen: three of them, backed by FastAPI and DuckDB services reading GeoParquet straight from S3. Scroll through them and the dashboard on the right becomes each app in turn.
App 1 of 3
Hydrotechnical dashboard
Every place a pipeline crosses a river gets a scenario: how high the water stands at mean annual flow, at the two-year flood and at the ten-year maximum, how deep the flow scours the bed, and how much cover that leaves over the pipe. This is one real crossing, a wide river in a deep valley. Keep scrolling and the water rises to the ten-year flood. The channel nearly triples in width, spills onto a floodplain where the pipe runs shallow, and about 350 feet of it lose their cover.
App 2 of 3
3D pipeline viewer
About 300 metres of line at that same crossing, as the viewer shows it: terrain coloured by height, a curtain hung along the pipe's route with the pipe drawn on it, so you can see how deep it runs under each rise. In the app, the modelled pipe is checked against depth-of-cover surveys along the same line. The viewer also sets models like this onto the map in their true position, so an engineer can fly from the whole network down to a single slope.
App 3 of 3
Channel migration
Rivers move. The channel-migration app traces each river from satellite imagery, year after year, and lays the traces over the pipelines, so an engineer can see where a channel is wandering long before a bank reaches the pipe. This is one bend where a river crosses a pipeline corridor, traced every year from 2016 to 2024. Scroll and the years stack up: the channel never sits in quite the same place, and the stretch of pipe it has crossed grows from about 500 feet to over 1,200.
Adgile Media: Trucks, Pings and Proof
The same months, a second clock: New York's billboard trucks.
At Adgile I built the Fleet Dashboard, the platform behind a fleet of mobile billboard trucks: live tracking, route history, and the geofences that prove to an advertiser their campaign was actually seen. Here is one truck's morning, the way the platform sees it.
Step 1 of 3
From pings to streets
Every truck reports where it is every few seconds, and the raw positions drift: a few metres most of the time, now and then a jump across the block. Each reading lands twice, in Firebase for the live map and in PostGIS for the record. Then a routing engine matches the trace to the streets actually driven and scores its own confidence, so a bad trace gets flagged instead of quietly trusted. That engine is expensive to run, so it starts only when there is matching to do and shuts itself down afterwards.
Step 2 of 3
Who could have seen it
A campaign is sold on eyes, not miles. Every day the audience files arrive in S3, and the platform joins them to the matched route, stretch by stretch. A loop through the busy core is worth far more than the same distance on quiet side streets, and the ribbon swells to show it.
Step 3 of 3
Proof for the client
The advertiser cares about one neighbourhood. Draw it as a geofence and PostGIS clips every route to it: the miles, the minutes and the impressions inside, ready to export as the PDF or spreadsheet the client receives. That export is the product. It is what turns a truck driving around into a campaign someone will pay for again.
Now: NYU CUSP and the CERA Lab
This fall I moved to Brooklyn, and from building tools to asking questions with them.
I'm doing a master's in Urban Data Science at NYU's Center for Urban Science + Progress. Alongside it I'm a graduate research assistant in the CERA Lab, working with Professors Augustin Guibaud and Yuki Miura on how cities adapt to wildfire: comparing sites in Los Angeles and Tulsa, harmonising satellite imagery with land-management records, and mapping destruction and rebuilding before and after a fire.
It's the same problem I spent three years on in industry, satellite data meeting messy administrative records, now pointed at a question about who recovers and how fast. The plates in the Atlas below come from this year.
Next: Datum, and the Questions Behind It
This chapter is a forecast, but part of it has already started.
Alongside school I'm building my own company. We go by the name Datum, after the surface every map is measured from, and we're making a suite of geospatial tools: GIS and photogrammetry tools, GeoAI, and civic tools that turn public data into something a resident or a planner can actually use. It's the kind of work I spent three years shipping for other people, now built on our own terms.
Research is the other half, and the part I'm most passionate about. I like the slow work of reading the literature until a gap shows itself, then going after it with data. The questions I keep coming back to are about urban equity, transit and who can get where, and the civic tools that could close those gaps.
If you're working on cities, transit or civic tech, we should talk.
The Atlas
Selected work, in two parts: research plates, and geoweb applications built for real users.
IResearch & analysis
- Plate I Measuring the rise of Hudson Yards with LiDAR
- Plate II Tracing the Canadian wildfire smoke over New York City
- Plate III A demolition on the dining table: change detection from phone video
- Plate IV Karachi, the city that stacked: population against building footprint
- Plate V Mapping the accessibility gap: subway elevators and the New Yorkers who need them
- Plate VI Heat in a desert city: Phoenix in the July 2023 heat record
IIGeoweb applications
- App I Hydrotechnical Dashboard: floods over buried pipe at river crossings
- App II 3D Pipeline Viewer: turning the ground over to see the pipe
- App III Channel Migration: years of river outlines over pipelines
- App IV Fleet Dashboard: tracking mobile billboards across US cities
- App V Recovery Resource Map: finding help in Wilkes County, North Carolina
Contact
The best way to reach me is a short note to yamama.amirr321@gmail.com. I read everything.