Automated Metadata Generation Enrichment

Using LLMs to automate the generation of metadata for items in libraries collections or enriching existing metadata has been identified as a promising use case.

• Given an image of a book's title page, generate title, author, and other properties as part of a FOLIO Instance JSON record
• From a prompt (experiment with Chatting with Catalogs use case 1), generate and validate a FOLIO Instance JSON record
• From a prompt, generate a FOLIO Holding and Item JSON records that are associated with an existing FOLIO Instance.
• From a FOLIO order record, generate a FOLIO Instance and Holding records

Stakeholders

Requirements

Possible Workflow

This diagram illustrates the expected workflow for using edge-ai, FOLIO, and the Apache Airflow workflow engine.

Metadata Generation from Textual Prompt

graph TD;
  FOLIO[FOLIO UI]-->eai[edge-ai API];
  nb[Notebook]-->eai[edge-ai API];
  cmdline[Command Line]-->eai[edge-ai API];
  speech[Spoken Audio]-->eai[edge-ai API];
  eai[edge-ai API]-->chgpt[Chat GPT];
  eai[edge-ai API]-->claude[Claude];
  eai[edge-ai API]-->goog[Google Gemini];
  eai[edge-ai API]-->local[Local Llama];
  eai[edge-ai API]-->whisper[Whisper];
  chgpt[Chat GPT]-->eai2[edge-ai API];
  claude[Claude]-->eai2[edge-ai API];
  goog[Google Gemini]-->eai2[edge-ai API];
  local[Local Llama]-->eai2[edge-ai API];
  whisper[Whisper]-->eai2[edge-ai API];
  eai2[edge-ai API]-->airflow1[Airflow DAG];
  airflow1[Airflow DAG setup]-->airflow2[Airflow Record Enhancement];
  airflow2[Airflow Record Enhancement]-->FOLIO1[FOLIO API];
  FOLIO1[FOLIO API]-->eai3[edge-ai API];
  eai3[edge-ai API]-->FOLIO2[FOLIO UI];
  eai3[edge-ai API]-->nb2[Notebook];
  eai3[edge-ai API]-->cmdline2[Command Line];

Loading

Metadata Generation from uploaded Images

graph TD;
  photo[Photos taken]-->FOLIO[FOLIO UI];
  photo[Photos taken]-->nb[Notebook];
  photo[Photos taken]-->cmdline[Command Line];
  app[Mobile App]-->eai[edge-ai API];
  FOLIO[FOLIO UI]-->eai[edge-ai API];
  nb[Notebook]-->eai[edge-ai API];
  cmdline[Command Line]-->eai[edge-ai API];
  eai[edge-ai API]-->chgpt[Chat GPT];
  eai[edge-ai API]-->claude[Claude];
  eai[edge-ai API]-->goog[Google Gemini];
  eai[edge-ai API]-->local[Local Llama];
  chgpt[Chat GPT]-->eai2[edge-ai API];
  claude[Claude]-->eai2[edge-ai API];
  goog[Google Gemini]-->eai2[edge-ai API];
  local[Local Llama]-->eai2[edge-ai API];
  eai2[edge-ai API]-->airflow1[Airflow DAG];
  airflow1[Airflow DAG setup]-->airflow2[Airflow Record Enhancement];
  airflow2[Airflow Record Enhancement]-->FOLIO1[FOLIO API];
  FOLIO1[FOLIO API]-->airflow3[Airflow Response];
  airflow3[Airflow Response]-->eai3[edge-ai API];
  eai3[edge-ai API]-->FOLIO2[FOLIO UI];
  eai3[edge-ai API]-->nb2[Notebook];
  eai3[edge-ai API]-->cmdline2[Command Line];

Loading

Screenshot of DAG

Resources

• Saccucci, C., & Salaba, A. (2021). Introduction to artificial intelligence (AI) and automated processes for subject access. Cataloging & Classification Quarterly, 59(8), 699-701.

• Owens, N., & Thomas, C. (2019). The future of cataloging in a FOLIO environment. The Serials Librarian, 76(1-4), 66-71.