Industry work · out-of-home advertising · US metros

Fleet Dashboard

Where every billboard truck is now, where it has been, and proof for the advertiser that it was there.

Company: Adgile Media Group, which runs advertising on trucks
Role: Geospatial software engineer, one of two engineers on the platform
Years: 2025–2026
Stack: React · Mapbox GL · NestJS · PostgreSQL + PostGIS · Firebase · Google Cloud

The live map: the app's sidebar, a list of truck cards with speed, mileage, campaign and market, and truck markers across the United States, grey when parked and blue on a live campaign

The brief

Advertisers pay to be seen, so prove it

Adgile wraps box trucks in a brand's advertising and drives them through the streets of US cities on planned routes. A GPS unit on each truck reports its position, speed, battery and signal all day long. On their own those pings are just noise; the business needs them answered as questions. Where is each truck right now? Did it drive the route the client paid for? How many people could have seen it, and where?

The dashboard is where staff ask those questions. I worked on it from the first commit, on both the web app and the API behind it, alongside one other engineer: the live map, route history, geofencing, the fleet health tab, the CRM sync and truck availability are the parts I built and describe here.

01 · Route history and impressions

Replay any week, for any truck

Pick a date range and any set of trucks, by name, campaign or city, and their routes come back snapped to the street network. Each dot is an impression point: a moment where a measurement partner counted the people who could see the truck, and every one opens a card with its numbers. Every query becomes a layer on the right, with its own colour and opacity, so several can be compared, hidden or exported.

Upload planned routes as GPX or CSV, several files at once, to lay the plan over what was driven
The map fits itself to wherever the trucks went, even when a campaign runs in several cities

Routes view over east Los Angeles: a week of routes for ten trucks drawn as chains of dark blue impression points along the streets, with the query panel on the left and the layer list on the right

02 · Geofencing

Draw a shape, get the campaign inside it

Clients often care about one neighbourhood, a stadium or a stretch of avenue, not the whole day's drive. Staff draw a polygon on the map and the API clips every road-matched route to it in PostGIS, in one query. Inside the shape, routes turn orange and GPS points red; outside they stay blue. A statistics table then totals the trucks, impressions and days of operation inside, ready to export as CSV or Excel.

Hand-drawn shapes cross themselves, so each is checked and repaired before it touches a query
Truck and campaign details are copied onto each route when it is written, so the clip needs no joins and stays fast
Exports stream row by row, so a whole campaign downloads instead of stopping at a row cap

A geofence polygon drawn over Monterey Park and Montebello: routes inside are orange with red GPS points, routes outside stay blue

Geofence statistics: totals for devices, impressions, daily average and operation days, a per-truck table, and CSV and Excel export buttons

try it · a geofence on made-up routes

	Inside	Of route	Impressions
Truck A
Truck B
Truck C
All

Try it

The same clip, in your hands

Drag the orange shape, or any of its corners. The three routes are cut to it the way the dashboard's query cuts real ones, and the table adds up the distance and impressions inside. The routes and numbers here are made up.

03 · Fleet health

Find the truck that went quiet

A truck with a flat tracker is a campaign with a hole in its proof. The health tab sorts every device into online, warning and offline, beside its last report, battery, signal, ignition and satellite count, so a dead unit can be caught while the campaign is still running. It splits the fleet into assigned devices, spare inventory and devices on live campaigns, searches across every field, and exports.

Device health: counts of online, offline and warning devices and a table of devices with status, last seen, battery, signal, ignition and satellites

Truck availability results for Los Angeles: 56 trucks available against 5 requested, marked fulfillable, with each truck's wrap dimensions, connectivity and 30-day impressions rank

04 · Truck availability

Can we take this campaign?

When a client asks for, say, five trucks in Los Angeles next week, the answer is spread across trucks and deals in the CRM. Sales enter the city, the dates and the number of trucks. The API takes every truck in that city marked ready to use, removes any truck already booked on a deal whose dates overlap, and says whether the request can be filled.

Ranked by recent reach: each free truck carries its impressions over the last 30 days
Wrap dimensions for every truck, side, rear and over the cab, beside the result

05 · Also in the app

Audits that keep the CRM honest

Two more tools in the same app, which my colleague led: an install check that lists every live campaign whose trucks have no installation booked, and a sync that keeps truck names in the CRM and the tracking system the same, logging every change and every failure.

Install checker: totals for deals, missing installs, missing trucks and covered trucks, and a list of campaigns with how many of their trucks are covered

Truck name sync history: a log of syncs with timestamps, success or error status, and the reason for each error

How it works

Four streams, one dashboard

Four kinds of data arrive at different speeds, from seconds to once a night. Follow a colour from where it starts on the left to the screen staff use on the right.

1Live position
2Route history
3Impressions
4Bookings

1Live position

Trucks→API→Firestore→Live map · Health

Each truck's latest state is pushed to every open browser.
2Route history

Trucks→API→PostGIS⇄OSRM→Routes

Every ping is kept, snapped to roads and served as map tiles.
3Impressions

Partner→API→PostGIS

Nightly files are matched to pings by truck and time.
4Bookings

HubSpot→API→Availability

Trucks and deals come from the CRM; campaign dates are copied onto routes.

Making it hold up

Months of pings, answered in seconds

A fleet reporting all day piles up rows quickly, and staff ask for whole campaigns at once. The slow requests were the ones people needed most, and a lot of my later work went into them: indexes the API checks for and creates when it starts, a time limit on heavy statistics queries so a bad request fails cleanly instead of hanging, and an Excel export that streams rows as they are read, so a campaign's raw pings download in full instead of stopping at a row cap.

Live mapweb app
Route history & GPXweb app
Geofencingweb app and API
Fleet healthweb app
HubSpot syncAPI
Truck availabilityweb app and API