AI & MLChatGPTOpen DataGovernment Data

Researching a Public Supply Chain with a ChatGPT Agent

A complex public project can involve many suppliers, systems, and countries. A question such as “how much of it is Canadian?” sounds simple until the unit of measurement needs to be defined.

My River-class destroyer project uses ChatGPT-assisted research to examine publicly described systems and suppliers. Its reports and visualizations are an independent research exercise, not an official bill of materials.

Organize the research before summarizing it

The retained materials cover component details, country breakdowns, a summary, and an initial cost exploration. The repository also includes NotebookLM presentation materials generated from those reports.

Separating the detailed reports from their visual summaries makes it possible to ask where a chart's conclusion came from. A presentation inherits the assumptions of its source reports, even when it makes the story easier to follow.

Define the denominator

The project's country percentages count identified major systems or supplier packages. They are not percentages of procurement spending, physical weight, labor, or economic value.

A large complex system and a much smaller equipment package can each count as one item. Without that explanation, a chart can imply far more than the underlying method supports.

Country assignment needs similar care. A supplier's headquarters, design origin, manufacturing location, and a best-supported estimate are different kinds of evidence. Combining them into one column requires keeping the classification rule visible.

What AI contributes

An agent can help organize a broad collection of public material, turn it into structured records, and draft explanations. The resulting records still need source review, particularly where a supplier relationship is inferred or a location is uncertain.

The most useful output is therefore a research trail: identified system, source, country-assignment basis, and uncertainty. A polished infographic is easier to read, but it should remain connected to that trail.

A reusable research pattern

This project is about AI-assisted research and communication, rather than a defence AI model. Its broader lesson applies to many supply-chain questions: decide what is being counted, keep estimates distinct from confirmed records, and do not let a summary outrun its sources.

I would update factual sourcing claims against official material before republishing a numerical conclusion. This article focuses on the method and the limits of the retained analysis.

Project material