How it Works
The Catalogue page
Our home page is the main catalogue. It presents the results of the latest search. You can reach it by clicking the main website title or ‘catalogue’ in the main menu. On opening, it’s displaying all the items in the collection.
- do a search – catalogue
- choose an item – item card
- see the metadata – item page
- open the document – PDF
Item cards and Item pages
In this collection, each document and its metadata (name, date, map descriptions, etc.) is called an ‘item‘. A search produces a page of item cards, much like catalogue cards in a bricks and mortar library. These results can be displayed in different ways, available from the search bar at top.
These link you through to the item pages, which display everything the librarian knows about a given document. Items in the collection can describe PDFs, graphic maps, spreadsheets, etc. In many cases they include the full text of the document. They also provide links to the documents themselves.
Search
When you use the search bar, the catalogue displays all the item cards that match your terms. The way you search is unique to you, in life and in the library.
In the main search window, if you put in ‘water’, you’ll collect about half of the documents we have. To narrow the search, you might use the filter by subjects & document-type in the left sidebar.
On any item page, if the full text field is open onscreen, you can search both document and metadata at the same time.
Advanced Search
Another approach is the tantalizing ‘advanced search‘. This gives you the power to pick one or more data bits (aka ‘metadatum’), like authors or source organizations, and search just in those fields. Maybe you know a person named Groucho who wrote a great water report. You’ll find it more easily by author.
Let’s say you want to find any maps that are inside docs in the collection. You can select the ‘map descriptions’ field and select ‘has value’. This just means that there is content in that field, indicating a map exists. You can narrow it down further. Click ‘add another search criterion’ and use ‘contains’ and the term ‘grafton’. You’ll find one or more maps of grafton lake. You can delete any search terms you don’t need anymore.
To AI or not to AI, is that the question?
The point of the site is to make relevant documents discoverable and provide them for study. It relies on traditional database software, and AI is not used for any of the actions described above. For the data, AI (using programmatic tools) was used to extract text and write metadata fields.
The collection is designed to be accessible to AI if you choose to use it, and it’s possible to ask questions comparing and analyzing documents. It may be possible to find common threads that provide new insights. To get your AI involved, you just:
a) do a search to surface documents you may want to analyze.
b) open each in a new tab. Copy the URL
c) paste the URL in a new document or directly in the chat window of your AI, with a suitable prompt.
The model will have access to the document as well as the library metadata about it. If you provide more than one link, they join the mix. Keep track of token usage, in case your model has context limits.
You can also:
a) generate a file that tells the AI which documents to go look at online, and/or
b) download the documents and metadata for analysis using your own computer, and/or
c) use offline AI models for privacy and better energy efficiency
The quality of AI analysis depends on the quality of the models and prompts used. Ask your AI to provide citations for its claims and always clearly indicate when it is imagining, extrapolating, guessing or inferring. Check the citations with the original documents before coming to any conclusions about their merit. There are many ways to get verifiable and accurate results with experience and well-crafted prompts.
How to point an AI model to a specific set of library items
Let’s say you have done an advanced search, and it returns 20 documents. What if you would like your AI model to come to this website and analyze that specific group only (saving many tokens). You could provide a URL for each item, but that’s a lot of copying and pasting. Instead you can:
1) perform a search
2) in the search bar, click on View As > Simple Json > dublin-core mapper > ‘open externally’ (see screenshot)
3) copy/paste the output into a chat window or save it as a .txt file and add that to the chat.
How this trick works
This exports a computer-readable file that includes just a portion of the metadata for the selected items, enough for the AI to find the documents you’re interested in without swallowing the entire database. You don’t need to read the file, just save it as a text document or copy/paste it into your chat window along with your prompt. When the AI visits the listed pages, it has access to all the metadata there, not just the info in the json file.
Json format can look like a foreign language, but if you just ignore that and copy/paste it, it soon stops looking strange.
One limitation of this method is that when you only want certain records from the search, you need to either edit the file or explain to the model which records you are interested in, and to ignore or edit out the others.
What is Tainacan?
Tainacan is a free, open-source Brazilian solution for managing and publishing digital collections. It transforms a WordPress site into a dynamic and customizable repository, ideal for cultural institutions, research projects, and digital collections of any kind. It is the main software driving the EcoLibrary, as well as most of the top museums in Brazil.
Main Tainacan English Website: tainacan.org/en/