Ollama for OCR, transcription and data extraction

This workshop will introduce the use of the Ollama application to extract text and structured data from images and sound recordings. It will cover installation and operation of Ollama, and will provide worked examples which can be explored in real time. Ollama provides a chat interface for familiar chat interactions, and also a network API which allows programmatic access to its service, using scripts in various programming languages (the examples will use Python). Ollama is free and open source, and can be installed on Windows, Mac and Linux machines.

It uses free downloadable LLMs which are useful for a variety of tasks such as OCR, speech transcription, or even deriving structured data such as CSV or JSON. With a good GPU and the use of Python scripts it is possible to automate these tasks to handle large sets of page images. The quality depends on the size of the model, which in turn depends on the hardware. The best results come when running on a machine with an NVIDIA GPU or a Mac with integrated memory; but even on a simpler system it can be used for private investigation of the capabilities of LLMs for various tasks.

The workshop will demonstrate three tasks: OCR of a handwritten text, transcription of a voice recording, and extraction of CSV data from a structured printed document. The examples will use the desktop application and Python scripts. It will familiarize you with the basic methods of applying LLMs to these research problems. The processes can be transferred to paid online AI services, but for most research purposes, output from a local LLM running on a reasonably capable workstation will be of equal quality.

Register here!

Instructor: Peter Binkley is a librarian with the U of A Library, where he offers workshops and consultations to support researchers with problems and methods of digital scholarship. He has worked in digitization, digital librarianship and digital preservation since the early years of this century. His academic background is in textual studies in the Latin Middle Ages, and he has a Ph.D. in Medieval Studies. His research is now mostly devoted to the 20th century history of pre-digital research technologies such as microfilm.