Tech Stack

Eco knowledge needs discoverability.

Every document welcomed into the collection goes on a journey. It turns from a shy file on an obscure flash drive to a fully visible, contributing member of our shared environmental knowledge.

1. We process a PDF report or other file with an advanced prompt using Claude Sonnet to extract about 35 fields of metadata, such as title, author, map descriptions, gis geometry etc. We look for errors and add details.
2. We feed the data into a museum database from Brazil called Tainacan. It sits on a WordPress website.
3. We post the original digital documents and text-only versions on the site as well. You can download the documents.
4. The document joins our collection of other items on the EcoLibrary website. You can discover them by searching on any of the metadata fields, by title, author, subject etc. They are discoverable by search engines.
5. We convert the same data to ’embeddings’ through AIEngine.
6. We also convert the original PDFs, museum database and the pages of this website to embeddings.
7. We store these in a vector database called Pinecone.
8. AIEngine provides the chat interface, currently using Claude Sonnet. It responds to your questions by searching for relevant information across the collection by meaning.
9. You can use the chat to ask questions with the whole library as your research engine. It provides citations with links and suggests followup questions.

Have fun playing around with the site. Please let us know how it feels and what you discover.

See the Contribute page for more information about our mission going forward.

Bowen Island Conservancy Tech Team