In today's fast-paced digital world, businesses are constantly seeking innovative ways to harness the power of their data. Papillon Solutions GmbH has taken a significant leap forward in this endeavor by enhancing its product, Boolee, through a strategic collaboration with SWISDATA, a nonprofit research organization. This partnership was made possible by leveraging the FFG Innovation Voucher, a program designed to foster innovation and technological advancement.

Boolee, an LLM-based chatbot, has been at the forefront of generating statistics from analytic data. However, its capabilities have now been expanded to handle custom data uploads, including various file formats such as CSV, Excel, and Parquet. This enhancement is made possible by a semantic layer developed by SWISDATA, which automatically creates descriptions from tabular data, enabling Boolee to understand and query diverse datasets effectively.

The collaboration resulted in two key deliverables: a comprehensive paper detailing the integration of large language models (LLMs) and AI methods, and a fully functional prototype. The prototype, built using TypeScript and featuring a command-line interface, leverages DuckDB for data loading and analysis. It examines data properties such as types, null values, distribution characteristics, and more, feeding these insights into LLM queries to identify column names, keys, and value usages.

At the heart of Boolee's enhanced capabilities lies a rule matching engine that allows predefined rules to process data inputs through computations or aggregations. Inputs are matched using semantic meaning and meta-information, with vector similarity generated by an open-source embedding model (bge-large). This ensures that uncertain matches are verified by LLMs, enabling recursive rule application for complex operations. The adaptability of this engine extends beyond the current application, making it a versatile tool for various domains.

The prototype utilizes self-hosted smaller LLM models, such as Llama 3.1 8B and QWEN 2.5 14B, which have proven to be effective without compromising performance. This approach not only avoids sending sensitive data to non-European cloud providers but also offers environmental benefits by reducing the carbon footprint associated with larger models. Additionally, smaller models are cost-effective and easier to manage, making them a sustainable choice for businesses.

In laboratory settings, the prototype has demonstrated exceptional performance, offering capabilities that are unmatched in the current market. By integrating LLMs with AI, analytic, and statistical methods, Boolee optimizes results across features to reduce bias and errors. This integration not only aids human operators but also enhances how systems understand data, making it an invaluable tool for organizations seeking to unlock deeper insights from their data.

Papillon Solutions GmbH's collaboration with SWISDATA, facilitated by the FFG Innovation Voucher, has led to significant advancements in Boolee. The integration of LLMs and a sophisticated rule matching engine positions Boolee as a cutting-edge solution for handling custom data uploads and generating meaningful statistics. With its focus on environmental sustainability, cost efficiency, and adaptability across domains, Boolee stands out as a revolutionary tool in the realm of data analytics.

Whether you're a business looking to enhance your data capabilities or simply curious about the future of analytics, Boolee offers a glimpse into what's possible when innovation meets practical application. Don't miss the opportunity to explore how Boolee can transform your approach to data—dive in and experience the power of next-generation analytics today!