On September 16, we hosted the final session of the Traceable AI Answers for Public Data project under the AI Learning Labs initiative. In collaboration with two national governments (Brazil and Uruguay), we have prototyped an AI integration with open data portals that enables users to ask natural-language questions and receive responses traceable directly to the underlying datasets.

If you missed the live session, you can now watch the full event recording and explore all presentation materials.

• 🎥 Watch the recording
• 🫆 Test the tool live

Our goal isn’t running isolated technical tests, but turning real-world trials into open blueprints that any government or civil society organisation can safely adapt.

Our partners from Uruguay (AGESIC) and Brazil (CGU) shared how they tested the framework on their internal data environments.

We publish a great deal of data on a wide range of topics, and we never know for certain how this data reaches the public. This pilot project has enabled us to take an important step in that direction.

— Gabriela Horta, National Directorate of Energy (DNE), Uruguay

“For us, the challenge going forward is to ensure it is a sustainable solution in terms of token consumption. We are testing it with various LLMs so that we can adapt the model to work with other important datasets.
— Gustavo Suarez, AGESIC, Uruguay

Being able to translate a complex topic such as parliamentary amendments into language that is accessible to the general public via an AI-powered chat interface is of enormous value.

— Otávio Neves, CGU, Brazil

Steven Flower from the International Aid Transparency Initiative (IATI) introduced why structured global data standards are ideal candidates for AI frameworks, setting up a live walk-through of our IATI Sandbox demo. 

For us in IATI, traceability is the real key thing, for example, on how we can follow the money from a government to an aid institution or partner. This project therefore speaks to the core of who we are and what we do.

— Steven Flower, IATI

At the heart of the session was the launch of our new publication, ‘Connecting AI to Government Data’.

The field guide synthesises everything we learned while implementing Model Context Protocol (MCP) frameworks alongside public institutions—offering practical guidance on data readiness, governance, technical architecture, and open replication.

• Download the Field Guide (PDF)
• Explore the MCP Technical Documentation

The Q&A segment sparked a great discussion around data governance, metadata standards, and technical architecture. A few key themes raised by the audience included:

Rethinking Metadata for AI

A key takeaway from the Field Guide: connecting AI to open datasets isn’t enough—it also needs context, clear domain definitions, and human experts to avoid misinterpretation. The audience took this further, suggesting data catalogs include sample questions or typical use-cases directly in their metadata.

Building Trust & Governance

Questions focused on how governments can safely deploy AI—including how to log system activity for transparency, route questions to the right departments, and choose the right AI models for data accuracy. 

With the pilot phase complete, we are using our pilot learnings to advance three main improvements:

• Strengthening the core standard: Feeding our real-world findings back into the MCP framework itself, so every future implementation benefits.
• Interactive data visualisations: Recent protocol extensions allow users to see interactive charts and graphs alongside text answers.
• Standardised source citations: We are introducing a formal proposal for standardised citations across AI tools, ensuring every AI-generated response links back cleanly to its original source data.

Thank you again to AGESIC, CGU, IATI, the Patrick J. McGovern Foundation, and everyone who joined us live!

Open Knowledge’s AI Learning Labs is an initiative that aims to experiment with AI, translate knowledge from social sector organisations around the world, and produce public, multilingual AI-literacy resources tailored for organisations addressing similar issues elsewhere.

Together, we will catalyse learning and develop replicable methods to help organisations build AI skills, use AI responsibly, and develop their own AI projects. All resources will be openly available at School of Data.

Join the conversation:

• Open Knowledge Forum
• School of Data community
• Contact our team: info@okfn.org

This project has been made possible thanks to the generous support of the Patrick J. McGovern Foundation (PJMF). We are grateful for our ongoing partnership in promoting digital literacy and investing in AI for the public good. Learn more about its funding programmes here.

Kannika Thaimai is a Project Manager at the Open Knowledge Foundation (OKFN). Bringing together people, ideas, and technologies, she builds collaborative programs that drive openness and social impact. For more than ten years, Kannika has managed cross-sector initiatives spanning public innovation, open-source ecosystems, and digital inclusion across Europe and East Africa. As part of the open knowledge movement, she founded and led UNLOCK, Wikimedia’s open innovation program empowering communities to develop tools and projects that advance free knowledge.