Transport Global Intelligence System — Concept & Direction

The Core Idea

Build a Global Intelligence System for Transport — an AI-powered layer that sits above the world's fragmented transport data platforms and lets anyone, regardless of technical skill, query, combine, and make sense of that data through natural language.

This is not a dashboard. It is not another database. It is the connective tissue between dozens of existing platforms, datasets, and knowledge sources that today require specialist skills to navigate individually and are almost never used together.

Why This Matters Now

Three forces have converged:

The UN Decade of Sustainable Transport (2026–2035) launched in December 2025 with an Implementation Plan that calls for progress monitoring across six priority areas. Proposals for a global tracking framework exist — most notably from SuM4All — but no integrated, operational system is in place to actually measure progress at scale. The data infrastructure to underpin the Decade's accountability goals remains largely unbuilt.
AI/LLM capabilities have reached the point where natural language querying of complex, heterogeneous data is technically feasible and affordable. This was not possible three years ago.
The data already exists — our audit of 25 sources found 14 with open REST/SDMX APIs requiring no authentication, and another 6 with free access. The problem is not data scarcity; it is data fragmentation.

What Makes This Different

There is no shortage of transport data tools. There are maps, dashboards, databases, and now even chatbots (GIZ's TDCI AI assistant). What none of them do is bridge across data types and sources.

The hard part is bridging data that was never meant to talk to each other:

Policy documents + GIS data: A question like "what road investments are planned in regions with the highest climate vulnerability?" requires combining narrative policy documents with spatial infrastructure data. Today that takes a specialist weeks. An AI agent could synthesise it in minutes.
Financial data + infrastructure data: Comparing cost-per-kilometre of tarred road across countries means joining IATI/DAC spending data with road network inventories. Nobody does this routinely.
Real-time monitoring + static indicators: Overlaying port disruption alerts (PortWatch) with trade dependency indicators (World Bank) gives early warning capability that neither dataset provides alone.

The principle: show that data which was never designed to be connected can be connected through AI, and that the combination yields answers no single platform can deliver.

Our Contribution

We come in as the how, not the what. The domain expertise lives with Oxford, UNECE, WRI, FCDO, GIZ, country governments and the wider coalition. Our contribution is:

AI architecture expertise — how do you make heterogeneous data AI-queryable? How do you combine unstructured text (policy docs) with structured data (indicators) with geospatial (maps)?
Data integration patterns — what does it mean to get data "AI-ready"? What are the practical steps from "25 data sources with different APIs" to "one conversational interface"?
Proof of concept — demonstrating the art of the possible with working examples, not slides
Honest brokerage — reality-checking what is feasible vs what sounds good in a pitch deck

We bring a toolkit and a way of working. The sector brings the knowledge, the data, and the users.

The Key Design Questions

1. Who is this for?

Two user archetypes:

Senior policymakers in partner governments — not tech-savvy, need plain-language answers to complex questions about transport investment, safety, infrastructure condition
Technical staff in road agencies, port authorities, rail authorities — more data-literate but siloed in their mode, currently navigating multiple disconnected platforms

A critical point from the Frontier Tech Hub: behaviour change. Giving people a tool does not mean they will use it. If they do not interrogate data today, a new tool alone will not change that. The system must be designed around how people actually make decisions, not around how we think they should.

2. Global or local?

Both. A transport minister in Ghana cares about Ghana, not global averages. A UNDESA analyst tracking the Decade needs the global view. The system must work at both scales — a "global-local intelligence system" that lets users zoom to their context while drawing on the full breadth of available data.

Most users will approach through a single mode (roads, maritime, rail, air). But the real value is in the connections between modes — port capacity affecting road freight costs, rail alternatives reducing road damage, climate vulnerability across networked infrastructure. The interface should allow mode-specific entry points while enabling cross-modal insight.

4. What are the first five data connections?

For any proof of concept, we need to start narrow and deep rather than broad and shallow. The initial selection should:

Cover different transport modes (not all roads)
Include different data types (GIS + structured + narrative)
Use the most accessible, well-documented sources
Demonstrate the principle of cross-source synthesis

From the data access audit, the most integration-ready candidates:

Source	Type	Mode Coverage	Access
OPSIS / open-gira	Geospatial infrastructure networks	Multi-modal (road, rail, air, maritime)	3 REST APIs, no auth
Transport Data Commons (UNECE)	Multiple datasets via single API	Multi-modal	Single API, emerging
World Bank Open Data	Indicators, LPI, spending	Cross-cutting	REST, no auth, CC-BY
PortWatch (IMF/Oxford)	Maritime trade monitoring	Maritime/ports	ArcGIS REST, no auth
IATI / DAC	Financial flows, aid spending	Cross-cutting	REST, no auth

With a policy document corpus (RAG) layered on top, this gives us structured data, geospatial data, financial data, and narrative knowledge — the full spectrum needed to demonstrate cross-type synthesis.

Two Ways of Working with Geospatial

A distinction that matters for the tool's design:

Querying maps — the user asks a natural language question; the AI agent queries geospatial data in the background and returns a text answer. ("Which East African road corridors have the highest climate risk?")
Showing maps — the user asks a question and the answer IS a map, with the right layers applied. The AI agent does not just interpret spatial data; it presents it visually.

Both are needed. The first is technically more straightforward (text-to-GIS-query). The second requires the system to generate or compose visual outputs, which is harder but often more impactful — sometimes a map with red-and-green corridors says more than a paragraph.

What We Are NOT Building

Not replacing any existing platform — we are a layer above them
Not a real-time control system — this is intelligence and planning
Not a finished product — this is proving a concept and building momentum
Not pretending to have domain expertise — we bring the AI and integration capability; the sector brings the knowledge

The Strategic Arc

Near-term (now through March 2026)

Landscape analysis and data audit (done)
Validate data priorities and fill gaps with domain experts (Oxford, FCDO, UNECE)
Build working examples of cross-source data synthesis
Aim toward something demonstrable for Transforming Transportation (second week of March 2026)

Medium-term (next financial year, from April 2026)

More capacity available for deeper work
Move from examples to a coherent proof-of-concept prototype
Refine user journeys based on feedback from the February/March events
Engage with WRI and other partners on sustained development

Long-term (aligned with the UN Decade, 2026-2035)

Position the system as the data backbone of the Decade
Expand from 5 to 50+ data sources
Build a federated, sustainable platform with proper governance
Move from FCDO seed funding toward institutional sustainability

Open Questions

What are the top priority data gaps that partners can help identify?
How does the GIZ TDCI AI assistant relate to this — complement or overlap?
What does IE Connect's real-time data collection methodology look like, and can it feed in?
Which OECD/DAC indicator set is FCDO working to update, and can it inform the taxonomy?
What is the right institutional home for something like this long-term?

This is a living document. It captures direction, not decisions. Updated as the work evolves.