Edge-AI and FOLIO at Stanford University Libraries
Prompt Cataloging Example
A user enters a text prompt and sends it through a Jupyter notebook, command-line interface, or user interface to the edge-ai API /inventory/instance/generate endpoint.
The edge-ai takes the prompt and using a ChatGPT model and a DSPy prompt signature, sends the prompt to the model to generate the beginning of a JSON Inventory Instance record.
We are using a FOLIO snapshot environment for demonstrating edge-ai
& ai-workflows in this and the next example.
The FOLIO snapshot environment:
The FOLIO snapshot environment:
- URL: folio-snapshot-2.dev.folio.org
- Login: diku_admin
- Password: admin
Edge-AI
Edge-AI Response
Apache Airflow Instance Generation DAG
Using the Apache Airflow REST API, edge-ai triggers an Airflow workflow or DAG (Directed Acyclic Graph).
The DAG then has different tasks to
- Retrieve the reference data containing UUIDs for data types like contributor, contributor name type, and instance type used in FOLIO
- Attempt to match the Instance Record to existing Instance records using RAG, and use a second prompt to the model to assign a match probability
- If a match is made, return the existing Instance UUID
- Or, create a new JSON Instance in FOLIO
