GPX: multi-tenant GPS tracking with AI-driven insight
GPX is a live GPS and BLE asset-tracking platform for operations and supply-chain teams. Built multi-tenant from day one on NestJS and PostgreSQL, it layers AI on top of raw location data so ops teams see risk and status, not just dots on a map.
·
Prefer WhatsApp or email?
The problem
Asset-tracking tools often ship single-tenant and get retrofitted for multiple customers later — a change that usually means rebuilding the data layer from scratch. The brief for GPX was to design for multiple customer organizations from the first schema, not after the first sales call asked for it.
On top of that, raw GPS/BLE location feeds are not useful to an ops team by themselves. The product needed to turn a stream of coordinates into something closer to "here is what needs attention," which meant AI had to sit on top of the tracking layer, not replace it.
What was built
A NestJS API on PostgreSQL with tenancy built into the data model from the start, so each customer organization’s assets, trackers, and users are isolated by design rather than filtered after the fact in application code.
Ingestion pipelines for GPS and BLE tracker data feeding a real-time visibility layer — dashboards and alerts that reflect current asset location and status rather than a stale daily snapshot.
An AI layer (OpenAI/LangChain-class tooling) on top of the tracking data for supply-chain insight — surfacing anomalies and risk signals from location and status patterns instead of leaving ops teams to eyeball a map all day.
Core features
- Multi-tenant data architecture
- GPS and BLE tracker ingestion
- Real-time asset visibility dashboards
- AI-driven anomaly and risk insight
- Role-based access per tenant
- Ops-focused alerting
Architecture & stack
- NestJS
- PostgreSQL
- React
- GPS/BLE tracker integration
- OpenAI / LangChain-class AI tooling
TODO(owner): add a simple architecture diagram asset for GPX if a screenshot is available.
Hard problems solved
Multi-tenancy as a day-one decision, not a retrofit
Adding tenant isolation after a single-customer schema already exists usually means a painful migration and a period of risky double-writes. GPX’s data model was designed with tenant boundaries from the first migration, which is slower on day one and considerably cheaper every day after.
Turning a location feed into something an ops team can act on
A map full of moving dots is not insight. The AI layer had to be scoped to a specific job — flag what is unusual or risky — rather than becoming a vague "AI-powered" feature with no clear output ops teams could act on.
Keeping real-time ingestion reliable across tracker types
GPS and BLE trackers behave differently — different update frequencies, different reliability characteristics, different failure modes when a device loses signal. The ingestion layer had to normalize both into one consistent asset-status model without losing the signal that makes each tracker type useful.
Outcome
GPX is live at gpx.co with multi-tenant tracking, real-time dashboards, and AI-driven insight working as one product.
TODO(owner): confirm any publishable tenant count, tracker volume, or customer numbers before adding metrics to this case study.
Client note
“GPS visibility plus an AI supply-chain story on gpx.co. Multi-tenant thinking from day one — the kind of proof UK and AU B2B buyers ask for.”
FAQ
Why build multi-tenant from the start instead of adding it later?
Retrofitting tenant isolation onto a single-customer schema usually means a risky data migration later. Designing tenant boundaries into the first schema costs more up front and far less over the product’s life.
What does the AI layer actually do in GPX?
It analyzes location and status data to surface anomalies and risk signals for ops teams — a specific, scoped job, not a vague "AI-powered" label with no defined output.
Does GPX support both GPS and BLE trackers?
Yes — the ingestion layer normalizes both tracker types into one consistent asset-status model so dashboards and alerts work the same regardless of hardware.
Can this architecture scale to more tenants later?
That is the point of designing tenancy into the schema from the start — adding new customer organizations does not require touching the core data model.
What would a comparable multi-tenant tracking SaaS cost?
Multi-tenant SaaS with real-time ingestion and AI features typically falls in the custom SaaS band from $4,999+ given the data-isolation and integration work involved. See the GPS asset tracking page for a scope breakdown.
Send a short brief
Building something in this space? Share goals, timeline, and budget band — you get a clear scope and USD quote.
Talk on WhatsApp / emailRelated: Matching hire page · Matching service page · add openai to existing saas · All work