The Waitangi Tribunal's record is one of the most significant collections of history we have in Aotearoa. The Tribunal’s inquiries and examinations into claims made by manawhenua, regarding breaches of the Te Tiriti o Waitangi (the Treaty of Waitangi), have resulted in over 75,000 publicly available documents.
These documents are crucial to our modern day decision making around the whenua (land), awa (rivers), and moana (sea) - be it ongoing treaty settlements, resource consents or governance arrangements. They are also supplemented by other contemporary hapū and iwi environmental and climate planning documents.
But too often this historical context, or even current context, is buried in a record that no one has time to search properly. Even with so many of these documents digitised, scanning 100GB of legislation, reports, briefs of evidence, and commissioned research could take days for a skilled researcher. For whānau, students, or anyone without the time and the training, the context could be out of reach entirely.
Nation Connect surfaces the same material in seconds. This greatly lowers the barrier of access. And while Nation Connect currently focuses only on the native Māori of Aotearoa New Zealand (tangata whenua), the challenge of surfacing relevant historical context of colonial land use decisions is one shared across indigenous peoples worldwide.
Our history informs our future. Nation Connect is about making sure that history is actually within reach. This is the vision of Merata Kawharu. With the support of Lincoln University/Te Whare Wānaka o Aoraki, Merata and MadeCurious have brought this vision to life: an AI initiative that empowers Māori, and eventually other indigenous peoples around the world, to gain easily navigable insights from hard to reach public archives.
So how does it achieve this?
How Nation Connect works
At its simplest, you ask Nation Connect a question the way you'd ask a person - "How did Crown forestry policy affect Māori land retention?" - and it answers in plain language, drawing on the actual Waitangi Tribunal documents.
Behind the simplicity is an approach called retrieval-augmented generation, or RAG. Generalised large language models answer from whatever they happened to absorb during their training. This is where the confidently wrong "facts" you may have heard about come from.
Nation Connect works the other way around. When you ask a question, it first searches the entire corpus of Tribunal documents for the passages most relevant to what you've asked. Only then does it hand those passages to the AI model and ask it to answer using that specific material. Every answer is built from real documents in the corpus, and every answer points back to them. You can open the sources and read them yourself. The tool does the finding, with the judgement maintained by the human doing the prompting.
Doing this across more than 75,000 documents needs some serious machinery underneath. The heavy lifting of indexing and searching the corpus runs on Google's Gemini File Search. It let us stand up search over the entire corpus with far less custom infrastructure than a system like this would traditionally demand, and at a fraction of the cost. The chat interface people actually use is built in Next.js, the service that ties it together and generates the responses is a Python application built on FastAPI, and the whole thing runs on Google Cloud Run so it scales with demand.
Plenty of this is off-the-shelf tech, and deliberately so - we prefer to build on proven, managed services than reinvent them. However, Gemini File Search is still an emerging technology. Getting it to work reliably across 75,000 documents still required some strong software engineering chops.
Ingestion of the entire corpus
Before any of this could work, the whole corpus had to be gathered, cleaned up, and loaded in. That meant scraping an index of every document on the Tribunal site, downloading the lot, and then wrestling it into a shape the tooling would accept.
Old Word and Excel files needed converting to modern formats first. The upload API is fussy about characters and won't take a filename longer than 40 of them, so we cleaned up every one. Anything over 100MB had to be split into smaller pieces. And the staging area only holds 20GB, so we were purging files constantly as they loaded, just to stay under the ceiling.
Then the upload itself: over 14 hours to bring the full corpus into the search store, so we really did not want it falling over in hour thirteen!
The initial Treaty of Waitangi corpus has been supplemented with additional public information, including hapū led environmental and climate planning documents.
Answers you can trace and trust
We wanted to ensure that any answers returned by Nation Connect can be validated. So a hard requirement from the start was that every response link back to the actual Tribunal document behind it, and open the original so you can read it in context.
That turned out to be one of the trickiest parts of the build. The document a user opens doesn't live inside Gemini File Search - it sits in Google Cloud Storage - so the two have to be tied together by an exact match on the file's name in each system. The metadata we'd hoped to use for the link, each document's original Tribunal URL, isn't handed back in the search results. Making the connection reliable meant very deliberate, consistent naming across both systems, and generating a secure link to the source on the fly every time someone opens one.
This traceability is key to building trust in AI-assisted tools, especially in such significant and sensitive kaupapa as this. The histories in these documents are contested and deeply held, and deciding what they mean belongs to the communities they belong to - to iwi, hapū and whānau. Pointing every answer back to its source keeps the tool in its place: a research aid, not an authority on the Treaty. It hands people the record, and leaves the meaning to them.
Current state and next steps
Nation Connect is live. This version 1 release is publicly accessible; ask it a question today. It has recently had more contemporary documents added, to expand on the Waitangi Tribunal corpus.
The most valuable thing right now is getting the tool in front of the people who'll actually use it, and listening to what they need.
Once Nation Connect is refined in the context of Aotearoa New Zealand, then the intention is to take it to other jurisdictions. The problem Nation Connect solves is not unique to the South Pacific. Indigenous peoples around the world share similar experiences of colonisation, with historical context and decisions hard to search and discover.
The technology underneath Nation Connect doesn't much care which archive it's pointed at. A grounded, traceable way to search a vast body of documents works just as well for another country's record as it does for the Waitangi Tribunal's. However, the Waitangi Tribunal is a massive asset that is not necessarily replicated in these other jurisdictions. And with so much indigenous historical context still shared via oral storytelling, different methods may be required to support the surfacing of this context.
Wherever Nation Connect goes, the kaupapa driving it will remain the same: empowering indigenous communities, and those working in support of them, with the history that informs and shapes their futures.
