Trustworthy AI at Scale: How Private LLMs Protect Data and Cut Costs

When building LLM applications, a significant portion of your time will be dedicated to data
wrangling – from content extraction and cleaning to de-duplication and filtering out problematic data. This talk introduces two open-source tools, Dockling and Data Prep Kit, designed to streamline these essential tasks. Attendees will learn how these tools can accelerate data preparation, improve overall data quality, and enhance the efficiency of building robust LLM applications.

Docling is a versatile document processor that handles various file types, including PDF, HTML, and DOCX. Data Prep Kit is a comprehensive Python library that democratizes and accelerates data preparation by providing out-of-the-box solutions for common tasks. Engineered to scale from a single laptop to large cloud clusters, it has been successfully used to process terabytes of data for training IBM Granite Large Language Models (LLMs).

Leveraging Docling for file parsing, Data Prep Kit offers a robust feature set including duplicate elimination, advanced document and code handling, language detection (for both spoken and programming languages), removal of personally identifiable information (PII), as well as spam, hate speech, and malware detection.

Join this session to explore how these tools can accelerate data preparation, enhance data quality, and improve the efficiency of building robust LLM applications. A live demonstration will also showcase some of their key features, time permitting.

More about docking: https://github.com/DS4SD/docling
More about Data Prep Kit : https://github.com/IBM/data-prep-kit