Skip to main content

A complete example

This example uses public Waitangi Tribunal material about the Wairarapa ki Tararua Report. It is deliberately modest. The purpose is to show the complete method with traceable public evidence, not to create a cultural place-name dataset from Tribunal material.

For a first wānanga run, use Te Manutukutuku Issue 64. It is a manageable PDF and pages 16–17 contain a short account of the Wairarapa ki Tararua inquiry. For a larger follow-up exercise, use the Waitangi Tribunal Wairarapa ki Tararua inquiry page and add one or more of the three full 2010 report volumes.

1. Keep the source locally​

Download the PDF into a source folder such as:

Local-AI-Demo\
  01-source\
    Te-Manutukutuku-Issue-64.pdf
  02-working\
  03-reviewed-data\

Keep the original filename or record the original URL. This is the evidence file that later results should point back to.

2. Check that the PDF has text​

Search the PDF for Wairarapa Moana. In the current official Issue 64 PDF, the page text is extractable without OCR. A direct text-extraction check on 22 August 2026 also preserved Māori text including Pouākani, Pāpāwai, Ōkautete and tāngata whenua.

This is worth checking before the AI step because a clean text layer removes one major source of error.

3. Add it to local AI​

Create an AnythingLLM workspace called something like Wairarapa public demo and add the PDF to the workspace.

Use a local language model and local embedding model. After the initial download and indexing, repeat the question with the network disconnected as described in Install local AI.

4. Ask a factual question first​

Try:

Using only this document, what does the article titled “The Wairarapa ki Tararua Report” say about Wairarapa Moana?

Give the page number and enough source wording for me to verify the answer.
Do not add outside information.

Then open the cited page. The point is to test retrieval before asking for a large structured extraction.

5. Extract candidate places​

Use this prompt:

Using only pages 16 and 17 of the article “The Wairarapa ki Tararua Report”, identify candidate geographic place names.

For each candidate return:
- place exactly as written
- likely geographic feature type only when supported by the text
- page
- a short context note
- confidence that the text is referring to a geographic place

Do not treat iwi, hapū or people as places merely because they have Māori names.
Preserve macrons exactly as supplied by the source.
Do not normalise spelling.
Do not invent coordinates.

A useful review should find obvious geographic candidates such as Cape Palliser, Wairarapa Moana, Pouākani, Te Ore Ore, Pāpāwai, Greytown and Ōkautete.

It should not convert Ngāti Kahungunu, Rangitāne, Te Hika-ō-Pāpāuma, Ngāi Tūmapuhia-ā-Rangi or Ngāti Hinewaka into place records merely because they appear in the same passage.

That distinction is one of the most useful checks in the demonstration.

6. Use a second page for historical naming​

Page 17 is especially useful because the article records a Tribunal recommendation that official place names be changed, including Rimutaka to Remutaka.

That gives a clear example of why the source form and the current geographic match need separate fields. The historical/source spelling remains evidence even after a modern Gazetteer match is found.

7. Verify selected public places​

For the downloadable demonstration dataset, three public candidates were checked against the current New Zealand Gazetteer on 22 August 2026.

Source formGazetteer matchName IDFeature typeGazetteer statusLatitudeLongitude
Cape PalliserCape Palliser / Mātakitaki-a-Kupe55640CapeOfficial Altered-41.612033175.273203
RimutakaRemutaka Range54796RangeOfficial Altered-41.270464175.092515
Wairarapa MoanaWairarapa Moana55638LakeOfficial Assigned-41.290448175.181891

The third record needs context. The Gazetteer description for Wairarapa Moana explains that the named feature is a collection of connected waterways including Lake Wairarapa, Lake Ōnoke and part of the Ruamahanga River. The review note in the demo data preserves that distinction rather than treating a single coordinate as the full cultural or geographic meaning of the moana.

The coordinates above come from the Gazetteer export, not from an LLM.

8. Download the reviewed demonstration data​

These files contain only the three public, traceable records above:

They are demonstration files, not an authoritative Māori cultural dataset. They retain the source wording, current public geographic match, Gazetteer Name ID and review notes.

9. Open it in QGIS​

For the CSV:

  1. Open QGIS.
  2. Add a delimited text layer.
  3. Use longitude as X and latitude as Y.
  4. Use EPSG:4326 for the Gazetteer latitude/longitude fields in this demo.
  5. Open the attribute table and compare place_as_written with matched_name.
  6. Inspect the three points against a suitable reference layer.

The point of the map is not that three points are visually impressive. It is that every point can be traced back through the geographic source to the document evidence.

See QGIS for the general application workflow.

10. Open it in Google Earth​

Open the KML in Google Earth. Each placemark includes provenance fields so the user can see that the geometry came from the New Zealand Gazetteer and the candidate name came from the Waitangi Tribunal source.

See Google Earth for the wider mapping guide.

11. Extend the exercise​

Once the small workflow is understood, replace Issue 64 with the full Wairarapa ki Tararua report volumes. Ask questions across several documents, then compare repeated references and alternative forms.

Do not increase the number of extracted names faster than you can review the evidence. A thousand candidate rows with weak provenance are less useful than a smaller table where every record can be checked.

What this example demonstrates​

The useful chain is:

public source → local copy → local retrieval → candidate name → source check → Gazetteer or appropriate authority → reviewed geometry → GIS

The local model makes document research faster. The source provides the evidence. The geographic source provides the public geometry. People remain responsible for interpretation, authority and decisions about use or publication.

Sources​

Source and Gazetteer records checked: 22 August 2026

Last reviewed: 26 August 2026

Last updated on